sentence1
stringlengths
52
3.87M
sentence2
stringlengths
1
47.2k
label
stringclasses
1 value
def create(self, name, plugin_name, hadoop_version, description=None, cluster_configs=None, node_groups=None, anti_affinity=None, net_id=None, default_image_id=None, use_autoconfig=None, shares=None, is_public=None, is_protected=None, domain_name=None): """Create a Cluster Template.""" data = { 'name': name, 'plugin_name': plugin_name, 'hadoop_version': hadoop_version, } return self._do_create(data, description, cluster_configs, node_groups, anti_affinity, net_id, default_image_id, use_autoconfig, shares, is_public, is_protected, domain_name)
Create a Cluster Template.
entailment
def update(self, cluster_template_id, name=NotUpdated, plugin_name=NotUpdated, plugin_version=NotUpdated, description=NotUpdated, cluster_configs=NotUpdated, node_groups=NotUpdated, anti_affinity=NotUpdated, net_id=NotUpdated, default_image_id=NotUpdated, use_autoconfig=NotUpdated, shares=NotUpdated, is_public=NotUpdated, is_protected=NotUpdated, domain_name=NotUpdated): """Update a Cluster Template.""" data = {} self._copy_if_updated(data, name=name, plugin_name=plugin_name, plugin_version=plugin_version, description=description, cluster_configs=cluster_configs, node_groups=node_groups, anti_affinity=anti_affinity, neutron_management_network=net_id, default_image_id=default_image_id, use_autoconfig=use_autoconfig, shares=shares, is_public=is_public, is_protected=is_protected, domain_name=domain_name) return self._patch('/cluster-templates/%s' % cluster_template_id, data, 'cluster_template')
Update a Cluster Template.
entailment
def get(self, cluster_id, show_progress=False): """Get information about a Cluster.""" url = ('/clusters/%(cluster_id)s?%(params)s' % {"cluster_id": cluster_id, "params": parse.urlencode({"show_progress": show_progress})}) return self._get(url, 'cluster')
Get information about a Cluster.
entailment
def update(self, cluster_id, name=NotUpdated, description=NotUpdated, is_public=NotUpdated, is_protected=NotUpdated, shares=NotUpdated): """Update a Cluster.""" data = {} self._copy_if_updated(data, name=name, description=description, is_public=is_public, is_protected=is_protected, shares=shares) return self._patch('/clusters/%s' % cluster_id, data)
Update a Cluster.
entailment
def verification_update(self, cluster_id, status): """Start a verification for a Cluster.""" data = {'verification': {'status': status}} return self._patch("/clusters/%s" % cluster_id, data)
Start a verification for a Cluster.
entailment
def create(self, name, plugin_name, plugin_version, cluster_template_id=None, default_image_id=None, is_transient=None, description=None, cluster_configs=None, node_groups=None, user_keypair_id=None, anti_affinity=None, net_id=None, count=None, use_autoconfig=None, shares=None, is_public=None, is_protected=None): """Launch a Cluster.""" data = { 'name': name, 'plugin_name': plugin_name, 'plugin_version': plugin_version, } return self._do_create(data, cluster_template_id, default_image_id, is_transient, description, cluster_configs, node_groups, user_keypair_id, anti_affinity, net_id, count, use_autoconfig, shares, is_public, is_protected, api_ver=2)
Launch a Cluster.
entailment
def _get_command(classes): """Associates each command class with command depending on setup.cfg """ commands = {} setup_file = os.path.join( os.path.abspath(os.path.join(os.path.dirname(__file__), '../..')), 'setup.cfg') for line in open(setup_file, 'r'): for cl in classes: if cl in line: commands[cl] = line.split(' = ')[0].strip().replace('_', ' ') return commands
Associates each command class with command depending on setup.cfg
entailment
def get_json(response): """Provide backward compatibility with old versions of requests library.""" json_field_or_function = getattr(response, 'json', None) if callable(json_field_or_function): return response.json() else: return jsonutils.loads(response.content)
Provide backward compatibility with old versions of requests library.
entailment
def create(self, name, data): """Create a Job Binary Internal. :param str data: raw data of script text """ return self._update('/job-binary-internals/%s' % urlparse.quote(name.encode('utf-8')), data, 'job_binary_internal', dump_json=False)
Create a Job Binary Internal. :param str data: raw data of script text
entailment
def update(self, job_binary_id, name=NotUpdated, is_public=NotUpdated, is_protected=NotUpdated): """Update a Job Binary Internal.""" data = {} self._copy_if_updated(data, name=name, is_public=is_public, is_protected=is_protected) return self._patch('/job-binary-internals/%s' % job_binary_id, data)
Update a Job Binary Internal.
entailment
def autoconvert(string): """Try to convert variables into datatypes.""" for fn in (boolify, int, float): try: return fn(string) except ValueError: pass return string
Try to convert variables into datatypes.
entailment
def call_plugin(plugin, f, *args, **kwargs): """Calls function f from plugin, returns None if plugin does not implement f.""" try: getattr(plugin, f) except AttributeError: return None if kwargs: getattr(plugin, f)( *args, **kwargs ) else: return getattr(plugin, f)( *args )
Calls function f from plugin, returns None if plugin does not implement f.
entailment
def create(self, job_id, cluster_id, input_id=None, output_id=None, configs=None, interface=None, is_public=None, is_protected=None): """Launch a Job.""" url = "/jobs/%s/execute" % job_id data = { "cluster_id": cluster_id, } self._copy_if_defined(data, input_id=input_id, output_id=output_id, job_configs=configs, interface=interface, is_public=is_public, is_protected=is_protected) return self._create(url, data, 'job_execution')
Launch a Job.
entailment
def update(self, obj_id, is_public=NotUpdated, is_protected=NotUpdated): """Update a Job Execution.""" data = {} self._copy_if_updated(data, is_public=is_public, is_protected=is_protected) return self._patch('/job-executions/%s' % obj_id, data)
Update a Job Execution.
entailment
def visitTerminal(self, ctx): """Converts case insensitive keywords and identifiers to lowercase Identifiers in quotes are not lowercased even though there is case sensitivity in quotes for identifiers, to prevent lowercasing quoted values. """ text = str(super().visitTerminal(ctx)) quotes = ["'", '"'] if not (text[0] in quotes and text[-1] in quotes): text = text.lower() return Terminal.from_text(text, ctx)
Converts case insensitive keywords and identifiers to lowercase Identifiers in quotes are not lowercased even though there is case sensitivity in quotes for identifiers, to prevent lowercasing quoted values.
entailment
def list(self, search_opts=None): """Get a list of Plugins.""" query = base.get_query_string(search_opts) return self._list('/plugins%s' % query, 'plugins')
Get a list of Plugins.
entailment
def convert_to_cluster_template(self, plugin_name, hadoop_version, template_name, filecontent): """Convert to cluster template Create Cluster Template directly, avoiding Cluster Template mechanism. """ resp = self.api.post('/plugins/%s/%s/convert-config/%s' % (plugin_name, hadoop_version, urlparse.quote(template_name)), data=filecontent) if resp.status_code != 202: raise RuntimeError('Failed to upload template file for plugin "%s"' ' and version "%s"' % (plugin_name, hadoop_version)) else: return base.get_json(resp)['cluster_template']
Convert to cluster template Create Cluster Template directly, avoiding Cluster Template mechanism.
entailment
def create(self, name, type, mains=None, libs=None, description=None, interface=None, is_public=None, is_protected=None): """Create a Job.""" data = { 'name': name, 'type': type } self._copy_if_defined(data, description=description, mains=mains, libs=libs, interface=interface, is_public=is_public, is_protected=is_protected) return self._create('/jobs', data, 'job')
Create a Job.
entailment
def list(self, search_opts=None, limit=None, marker=None, sort_by=None, reverse=None): """Get a list of Jobs.""" query = base.get_query_string(search_opts, limit=limit, marker=marker, sort_by=sort_by, reverse=reverse) url = "/jobs%s" % query return self._page(url, 'jobs', limit)
Get a list of Jobs.
entailment
def update(self, job_id, name=NotUpdated, description=NotUpdated, is_public=NotUpdated, is_protected=NotUpdated): """Update a Job.""" data = {} self._copy_if_updated(data, name=name, description=description, is_public=is_public, is_protected=is_protected) return self._patch('/jobs/%s' % job_id, data)
Update a Job.
entailment
def _query_for_reverse_geocoding(lat, lng): """ Given a lat & lng, what's the string search query. If the API changes, change this function. Only for internal use. """ # have to do some stupid f/Decimal/str stuff to (a) ensure we get as much # decimal places as the user already specified and (b) to ensure we don't # get e-5 stuff return "{0:f},{1:f}".format(Decimal(str(lat)), Decimal(str(lng)))
Given a lat & lng, what's the string search query. If the API changes, change this function. Only for internal use.
entailment
def floatify_latlng(input_value): """ Work around a JSON dict with string, not float, lat/lngs. Given anything (list/dict/etc) it will return that thing again, *but* any dict (at any level) that has only 2 elements lat & lng, will be replaced with the lat & lng turned into floats. If the API returns the lat/lng as strings, and not numbers, then this function will 'clean them up' to be floats. """ if isinstance(input_value, collections.Mapping): if len(input_value) == 2 and sorted(input_value.keys()) == ['lat', 'lng']: # This dict has only 2 keys 'lat' & 'lon' return {'lat': float_if_float(input_value["lat"]), 'lng': float_if_float(input_value["lng"])} else: return dict((key, floatify_latlng(value)) for key, value in input_value.items()) elif isinstance(input_value, collections.MutableSequence): return [floatify_latlng(x) for x in input_value] else: return input_value
Work around a JSON dict with string, not float, lat/lngs. Given anything (list/dict/etc) it will return that thing again, *but* any dict (at any level) that has only 2 elements lat & lng, will be replaced with the lat & lng turned into floats. If the API returns the lat/lng as strings, and not numbers, then this function will 'clean them up' to be floats.
entailment
def geocode(self, query, **kwargs): """ Given a string to search for, return the results from OpenCage's Geocoder. :param string query: String to search for :returns: Dict results :raises InvalidInputError: if the query string is not a unicode string :raises RateLimitExceededError: if you have exceeded the number of queries you can make. Exception says when you can try again :raises UnknownError: if something goes wrong with the OpenCage API """ if six.PY2: # py3 doesn't have unicode() function, and instead we check the text_type later try: query = unicode(query) except UnicodeDecodeError: raise InvalidInputError(bad_value=query) if not isinstance(query, six.text_type): raise InvalidInputError(bad_value=query) data = { 'q': query, 'key': self.key } # Add user parameters data.update(kwargs) url = self.url response = requests.get(url, params=data) if (response.status_code == 402 or response.status_code == 429): # Rate limit exceeded reset_time = datetime.utcfromtimestamp(response.json()['rate']['reset']) raise RateLimitExceededError(reset_to=int(response.json()['rate']['limit']), reset_time=reset_time) elif response.status_code == 500: raise UnknownError("500 status code from API") try: response_json = response.json() except ValueError: raise UnknownError("Non-JSON result from server") if 'results' not in response_json: raise UnknownError("JSON from API doesn't have a 'results' key") return floatify_latlng(response_json['results'])
Given a string to search for, return the results from OpenCage's Geocoder. :param string query: String to search for :returns: Dict results :raises InvalidInputError: if the query string is not a unicode string :raises RateLimitExceededError: if you have exceeded the number of queries you can make. Exception says when you can try again :raises UnknownError: if something goes wrong with the OpenCage API
entailment
def reverse_geocode(self, lat, lng, **kwargs): """ Given a latitude & longitude, return an address for that point from OpenCage's Geocoder. :param lat: Latitude :param lng: Longitude :return: Results from OpenCageData :rtype: dict :raises RateLimitExceededError: if you have exceeded the number of queries you can make. Exception says when you can try again :raises UnknownError: if something goes wrong with the OpenCage API """ return self.geocode(_query_for_reverse_geocoding(lat, lng), **kwargs)
Given a latitude & longitude, return an address for that point from OpenCage's Geocoder. :param lat: Latitude :param lng: Longitude :return: Results from OpenCageData :rtype: dict :raises RateLimitExceededError: if you have exceeded the number of queries you can make. Exception says when you can try again :raises UnknownError: if something goes wrong with the OpenCage API
entailment
def acquire(self): ''' Get a new connection from the pool. This will return an existing connection, if one is available in the pool, or create a new connection. .. warning:: If the pool was created with `maxsize` and `block=True`, this method may block until a connection is available in the pool. ''' self._condition.acquire() try: # Wait for a connection if there is an upper bound to the pool. if self._maxsize is not None and self._block: while not self._pool and self._nconnections == self._maxsize: self._condition.wait(timeout=None) # block indefinitely # Check the pool for a non-stale connection. while self._pool: pooledconn = self._pool.pop(0) # get least recently used connection if self._idlettl is not None and (pooledconn.released + self._idlettl) < time.time(): pooledconn.connection.close() self._nconnections -= 1 else: return pooledconn.connection connection = self._dbapi2.connect(*(), **self._connection_args.copy()) self._nconnections += 1 return connection finally: self._condition.release()
Get a new connection from the pool. This will return an existing connection, if one is available in the pool, or create a new connection. .. warning:: If the pool was created with `maxsize` and `block=True`, this method may block until a connection is available in the pool.
entailment
def release(self, connection): ''' Return a connection back to the pool. Prior to release, :py:meth:`ctds.Connection.rollback()` is called to rollback any pending transaction. .. note:: This must be called once for every successful call to :py:meth:`.acquire()`. :param connection: The connection object returned by :py:meth:`.acquire()`. ''' try: # Rollback the existing connection, closing on failure. connection.rollback() except self._dbapi2.Error: self._close(connection) return self._condition.acquire() try: if self._maxsize is None or self._maxsize > len(self._pool): self._pool.append(PooledConnection(connection, time.time())) self._condition.notify() else: self._close(connection) finally: self._condition.release()
Return a connection back to the pool. Prior to release, :py:meth:`ctds.Connection.rollback()` is called to rollback any pending transaction. .. note:: This must be called once for every successful call to :py:meth:`.acquire()`. :param connection: The connection object returned by :py:meth:`.acquire()`.
entailment
def finalize(self): ''' Release all connections contained in the pool. .. note:: This should be called to cleanly shutdown the pool, i.e. on process exit. ''' self._condition.acquire() try: if self._nconnections != len(self._pool): warnings.warn('finalize() called with unreleased connections', RuntimeWarning, 2) while self._pool: self._close(self._pool.pop().connection) self._nconnections = 0 finally: self._condition.release()
Release all connections contained in the pool. .. note:: This should be called to cleanly shutdown the pool, i.e. on process exit.
entailment
def _dict_merge(dct, merge_dct): """Recursive dict merge. Inspired by :meth:``dict.update()``, instead of updating only top-level keys, dict_merge recurses down into dicts nested to an arbitrary depth, updating keys. The ``merge_dct`` is merged into ``dct``. From https://gist.github.com/angstwad/bf22d1822c38a92ec0a9 Arguments: dct: dict onto which the merge is executed merge_dct: dct merged into dct """ for k, v in merge_dct.items(): if (k in dct and isinstance(dct[k], dict) and isinstance(merge_dct[k], collections.Mapping)): _dict_merge(dct[k], merge_dct[k]) else: dct[k] = merge_dct[k]
Recursive dict merge. Inspired by :meth:``dict.update()``, instead of updating only top-level keys, dict_merge recurses down into dicts nested to an arbitrary depth, updating keys. The ``merge_dct`` is merged into ``dct``. From https://gist.github.com/angstwad/bf22d1822c38a92ec0a9 Arguments: dct: dict onto which the merge is executed merge_dct: dct merged into dct
entailment
def _parse_schema(schema, method): """ Convert a Schema Object to a Python object. Args: schema: An ``OrderedDict`` representing the schema object. """ if method and schema.get('readOnly', False): return _READONLY_PROPERTY # allOf: Must be valid against all of the subschemas if 'allOf' in schema: schema_ = copy.deepcopy(schema['allOf'][0]) for x in schema['allOf'][1:]: _dict_merge(schema_, x) return _parse_schema(schema_, method) # anyOf: Must be valid against any of the subschemas # TODO(stephenfin): Handle anyOf # oneOf: Must be valid against exactly one of the subschemas if 'oneOf' in schema: # we only show the first one since we can't show everything return _parse_schema(schema['oneOf'][0], method) if 'enum' in schema: # we only show the first one since we can't show everything return schema['enum'][0] schema_type = schema.get('type', 'object') if schema_type == 'array': # special case oneOf so that we can show examples for all possible # combinations if 'oneOf' in schema['items']: return [ _parse_schema(x, method) for x in schema['items']['oneOf']] return [_parse_schema(schema['items'], method)] if schema_type == 'object': if method and all(v.get('readOnly', False) for v in schema['properties'].values()): return _READONLY_PROPERTY results = [] for name, prop in schema.get('properties', {}).items(): result = _parse_schema(prop, method) if result != _READONLY_PROPERTY: results.append((name, result)) return collections.OrderedDict(results) if (schema_type, schema.get('format')) in _TYPE_MAPPING: return _TYPE_MAPPING[(schema_type, schema.get('format'))] return _TYPE_MAPPING[(schema_type, None)]
Convert a Schema Object to a Python object. Args: schema: An ``OrderedDict`` representing the schema object.
entailment
def _example(media_type_objects, method=None, endpoint=None, status=None, nb_indent=0): """ Format examples in `Media Type Object` openapi v3 to HTTP request or HTTP response example. If method and endpoint is provided, this fonction prints a request example else status should be provided to print a response example. Arguments: media_type_objects (Dict[str, Dict]): Dict containing Media Type Objects. method: The HTTP method to use in example. endpoint: The HTTP route to use in example. status: The HTTP status to use in example. """ indent = ' ' extra_indent = indent * nb_indent if method is not None: method = method.upper() else: try: # one of possible values for status might be 'default'. # in the case, just fallback to '-' status_text = http_status_codes[int(status)] except (ValueError, KeyError): status_text = '-' for content_type, content in media_type_objects.items(): examples = content.get('examples') example = content.get('example') if examples is None: examples = {} if not example: if content_type != 'application/json': LOG.info('skipping non-JSON example generation.') continue example = _parse_schema(content['schema'], method=method) if method is None: examples['Example response'] = { 'value': example, } else: examples['Example request'] = { 'value': example, } for example in examples.values(): if not isinstance(example['value'], six.string_types): example['value'] = json.dumps( example['value'], indent=4, separators=(',', ': ')) for example_name, example in examples.items(): if 'summary' in example: example_title = '{example_name} - {example[summary]}'.format( **locals()) else: example_title = example_name yield '' yield '{extra_indent}**{example_title}:**'.format(**locals()) yield '' yield '{extra_indent}.. sourcecode:: http'.format(**locals()) yield '' # Print http request example if method: yield '{extra_indent}{indent}{method} {endpoint} HTTP/1.1' \ .format(**locals()) yield '{extra_indent}{indent}Host: example.com' \ .format(**locals()) yield '{extra_indent}{indent}Content-Type: {content_type}' \ .format(**locals()) # Print http response example else: yield '{extra_indent}{indent}HTTP/1.1 {status} {status_text}' \ .format(**locals()) yield '{extra_indent}{indent}Content-Type: {content_type}' \ .format(**locals()) yield '' for example_line in example['value'].splitlines(): yield '{extra_indent}{indent}{example_line}'.format(**locals()) yield ''
Format examples in `Media Type Object` openapi v3 to HTTP request or HTTP response example. If method and endpoint is provided, this fonction prints a request example else status should be provided to print a response example. Arguments: media_type_objects (Dict[str, Dict]): Dict containing Media Type Objects. method: The HTTP method to use in example. endpoint: The HTTP route to use in example. status: The HTTP status to use in example.
entailment
def _resolve_refs(uri, spec): """Resolve JSON references in a given dictionary. OpenAPI spec may contain JSON references to its nodes or external sources, so any attempt to rely that there's some expected attribute in the spec may fail. So we need to resolve JSON references before we use it (i.e. replace with referenced object). For details see: https://tools.ietf.org/html/draft-pbryan-zyp-json-ref-02 The input spec is modified in-place despite being returned from the function. """ resolver = jsonschema.RefResolver(uri, spec) def _do_resolve(node): if isinstance(node, collections.Mapping) and '$ref' in node: with resolver.resolving(node['$ref']) as resolved: return resolved elif isinstance(node, collections.Mapping): for k, v in node.items(): node[k] = _do_resolve(v) elif isinstance(node, (list, tuple)): for i in range(len(node)): node[i] = _do_resolve(node[i]) return node return _do_resolve(spec)
Resolve JSON references in a given dictionary. OpenAPI spec may contain JSON references to its nodes or external sources, so any attempt to rely that there's some expected attribute in the spec may fail. So we need to resolve JSON references before we use it (i.e. replace with referenced object). For details see: https://tools.ietf.org/html/draft-pbryan-zyp-json-ref-02 The input spec is modified in-place despite being returned from the function.
entailment
def convert_double_to_two_registers(doubleValue): """ Convert 32 Bit Value to two 16 Bit Value to send as Modbus Registers doubleValue: Value to be converted return: 16 Bit Register values int[] """ myList = list() myList.append(int(doubleValue & 0x0000FFFF)) #Append Least Significant Word myList.append(int((doubleValue & 0xFFFF0000)>>16)) #Append Most Significant Word return myList
Convert 32 Bit Value to two 16 Bit Value to send as Modbus Registers doubleValue: Value to be converted return: 16 Bit Register values int[]
entailment
def convert_float_to_two_registers(floatValue): """ Convert 32 Bit real Value to two 16 Bit Value to send as Modbus Registers floatValue: Value to be converted return: 16 Bit Register values int[] """ myList = list() s = bytearray(struct.pack('<f', floatValue) ) #little endian myList.append(s[0] | (s[1]<<8)) #Append Least Significant Word myList.append(s[2] | (s[3]<<8)) #Append Most Significant Word return myList
Convert 32 Bit real Value to two 16 Bit Value to send as Modbus Registers floatValue: Value to be converted return: 16 Bit Register values int[]
entailment
def convert_registers_to_float(registers): """ Convert two 16 Bit Registers to 32 Bit real value - Used to receive float values from Modbus (Modbus Registers are 16 Bit long) registers: 16 Bit Registers return: 32 bit value real """ b = bytearray(4) b [0] = registers[0] & 0xff b [1] = (registers[0] & 0xff00)>>8 b [2] = (registers[1] & 0xff) b [3] = (registers[1] & 0xff00)>>8 returnValue = struct.unpack('<f', b) #little Endian return returnValue
Convert two 16 Bit Registers to 32 Bit real value - Used to receive float values from Modbus (Modbus Registers are 16 Bit long) registers: 16 Bit Registers return: 32 bit value real
entailment
def connect(self): """ Connects to a Modbus-TCP Server or a Modbus-RTU Slave with the given Parameters """ if (self.__ser is not None): serial = importlib.import_module("serial") if self.__stopbits == 0: self.__ser.stopbits = serial.STOPBITS_ONE elif self.__stopbits == 1: self.__ser.stopbits = serial.STOPBITS_TWO elif self.__stopbits == 2: self.__ser.stopbits = serial.STOPBITS_ONE_POINT_FIVE if self.__parity == 0: self.__ser.parity = serial.PARITY_EVEN elif self.__parity == 1: self.__ser.parity = serial.PARITY_ODD elif self.__parity == 2: self.__ser.parity = serial.PARITY_NONE self.__ser = serial.Serial(self.serialPort, self.__baudrate, timeout=self.__timeout, parity=self.__ser.parity, stopbits=self.__ser.stopbits, xonxoff=0, rtscts=0) self.__ser.writeTimeout = self.__timeout #print (self.ser) if (self.__tcpClientSocket is not None): self.__tcpClientSocket.settimeout(5) self.__tcpClientSocket.connect((self.__ipAddress, self.__port)) self.__connected = True self.__thread = threading.Thread(target=self.__listen, args=()) self.__thread.start()
Connects to a Modbus-TCP Server or a Modbus-RTU Slave with the given Parameters
entailment
def close(self): """ Closes Serial port, or TCP-Socket connection """ if (self.__ser is not None): self.__ser.close() if (self.__tcpClientSocket is not None): self.__stoplistening = True self.__tcpClientSocket.shutdown(socket.SHUT_RDWR) self.__tcpClientSocket.close() self.__connected = False
Closes Serial port, or TCP-Socket connection
entailment
def read_discreteinputs(self, starting_address, quantity): """ Read Discrete Inputs from Master device (Function code 2) starting_address: First discrete input to be read quantity: Numer of discrete Inputs to be read returns: Boolean Array [0..quantity-1] which contains the discrete Inputs """ self.__transactionIdentifier+=1 if (self.__ser is not None): if (self.__ser.closed): raise Exception.SerialPortNotOpenedException("serial port not opened") if ((starting_address > 65535) | (quantity >2000)): raise ValueError("Starting address must be 0 - 65535; quantity must be 0 - 2000"); function_code = 2 length = 6; transaction_identifier_lsb = self.__transactionIdentifier & 0xFF transaction_identifier_msb = ((self.__transactionIdentifier & 0xFF00) >> 8) length_lsb = length&0xFF length_msb = (length&0xFF00) >> 8 starting_address_lsb = starting_address & 0xFF starting_address_msb = (starting_address & 0xFF00) >> 8 quantity_lsb = quantity & 0xFF quantity_msb = (quantity & 0xFF00) >> 8 if (self.__ser is not None): data = bytearray([self.__unitIdentifier, function_code, starting_address_msb, starting_address_lsb, quantity_msb, quantity_lsb, 0, 0]) crc = self.__calculateCRC(data, len(data)-2, 0) crcLSB = crc&0xFF crcMSB = (crc&0xFF00) >> 8 data[6] = crcLSB data[7] = crcMSB self.__ser.write(data) if (quantity % 8 != 0): bytes_to_read = 6+int(quantity/8) else: bytes_to_read = 5+int(quantity/8) data = self.__ser.read(bytes_to_read) b=bytearray(data) data = b if (len(data) < bytes_to_read): raise Exceptions.TimeoutError('Read timeout Exception') if ((data[1] == 0x82) & (data[2] == 0x01)): raise Exceptions.function_codeNotSupportedException("Function code not supported by master"); if ((data[1] == 0x82) & (data[2] == 0x02)): raise Exceptions.starting_addressInvalidException("Starting address invalid or starting address + quantity invalid"); if ((data[1] == 0x82) & (data[2] == 0x03)): raise Exceptions.QuantityInvalidException("quantity invalid"); if ((data[1] == 0x82) & (data[2] == 0x04)): raise Exceptions.ModbusException("error reading"); crc = self.__calculateCRC(data, len(data) - 2, 0) crcLSB = crc&0xFF crcMSB = (crc&0xFF00) >> 8 if ((crcLSB != data[len(data)-2]) & (crcMSB != data[len(data)-1])): raise Exceptions.CRCCheckFailedException("CRC check failed"); myList = list() for i in range(0, quantity): myList.append(bool((data[int(i/8)+3] >> int(i%8)) & 0x1)) return myList else: protocolIdentifierLSB = 0x00; protocolIdentifierMSB = 0x00; length_lsb = 0x06; length_msb = 0x00; data = bytearray([transaction_identifier_msb, transaction_identifier_lsb, protocolIdentifierMSB, protocolIdentifierLSB, length_msb, length_lsb, self.__unitIdentifier, function_code, starting_address_msb, starting_address_lsb, quantity_msb, quantity_lsb]) self.__tcpClientSocket.send(data) self.__receivedata = bytearray() if (quantity % 8 != 0): bytes_to_read = 9+int(quantity/8) else: bytes_to_read = 8+int(quantity/8) try: while (len(self.__receivedata) == 0): pass except Exception: raise Exception('Read Timeout') data = bytearray(self.__receivedata) if ((data[1 + 6] == 0x82) & (data[2 + 6] == 0x01)): raise Exceptions.function_codeNotSupportedException("Function code not supported by master"); if ((data[1 + 6] == 0x82) & (data[2+6] == 0x02)): raise Exceptions.starting_addressInvalidException("Starting address invalid or starting address + quantity invalid"); if ((data[1 + 6] == 0x82) & (data[2+6] == 0x03)): raise Exceptions.QuantityInvalidException("quantity invalid"); if ((data[1 + 6] == 0x82) & (data[2+6] == 0x04)): raise Exceptions.ModbusException("error reading"); myList = list() for i in range(0, quantity): myList.append(bool((data[int(i/8)+3+6] >> int(i%8)) & 0x1)) return myList
Read Discrete Inputs from Master device (Function code 2) starting_address: First discrete input to be read quantity: Numer of discrete Inputs to be read returns: Boolean Array [0..quantity-1] which contains the discrete Inputs
entailment
def write_single_coil(self, starting_address, value): """ Write single Coil to Master device (Function code 5) starting_address: Coil to be written value: Coil Value to be written """ self.__transactionIdentifier+=1 if (self.__ser is not None): if (self.__ser.closed): raise Exception.SerialPortNotOpenedException("serial port not opened") function_code = 5 length = 6; transaction_identifier_lsb = self.__transactionIdentifier&0xFF transaction_identifier_msb = ((self.__transactionIdentifier&0xFF00) >> 8) length_lsb = length&0xFF length_msb = (length&0xFF00) >> 8 starting_address_lsb = starting_address&0xFF starting_address_msb = (starting_address&0xFF00) >> 8 if value: valueLSB = 0x00 valueMSB = (0xFF00) >> 8 else: valueLSB = 0x00 valueMSB = (0x00) >> 8 if (self.__ser is not None): data = bytearray([self.__unitIdentifier, function_code, starting_address_msb, starting_address_lsb, valueMSB, valueLSB, 0, 0]) crc = self.__calculateCRC(data, len(data)-2, 0) crcLSB = crc&0xFF crcMSB = (crc&0xFF00) >> 8 data[6] = crcLSB data[7] = crcMSB self.__ser.write(data) bytes_to_read = 8 data = self.__ser.read(bytes_to_read) b=bytearray(data) data = b if (len(data) < bytes_to_read): raise Exceptions.TimeoutError('Read timeout Exception') if ((data[1] == 0x85) & (data[2] == 0x01)): raise Exceptions.function_codeNotSupportedException("Function code not supported by master"); if ((data[1] == 0x85) & (data[2] == 0x02)): raise Exceptions.starting_addressInvalidException("Address invalid"); if ((data[1] == 0x85) & (data[2] == 0x03)): raise Exceptions.QuantityInvalidException("Value invalid"); if ((data[1] == 0x85) & (data[2] == 0x04)): raise Exceptions.ModbusException("error reading"); crc = self.__calculateCRC(data, len(data) - 2, 0) crcLSB = crc&0xFF crcMSB = (crc&0xFF00) >> 8 if ((crcLSB != data[len(data)-2]) & (crcMSB != data[len(data)-1])): raise Exceptions.CRCCheckFailedException("CRC check failed"); if data[1] == self.__unitIdentifier: return True else: return False else: protocolIdentifierLSB = 0x00; protocolIdentifierMSB = 0x00; length_lsb = 0x06; length_msb = 0x00; data = bytearray([transaction_identifier_msb, transaction_identifier_lsb, protocolIdentifierMSB, protocolIdentifierLSB, length_msb, length_lsb, self.__unitIdentifier, function_code, starting_address_msb, starting_address_lsb, valueMSB, valueLSB]) self.__tcpClientSocket.send(data) bytes_to_read = 12 self.__receivedata = bytearray() try: while (len(self.__receivedata) == 0): pass except Exception: raise Exception('Read Timeout') data = bytearray(self.__receivedata) if ((data[1+6] == 0x85) & (data[2+6] == 0x01)): raise Exceptions.function_codeNotSupportedException("Function code not supported by master"); if ((data[1+6] == 0x85) & (data[2+6] == 0x02)): raise Exceptions.starting_addressInvalidException("Address invalid"); if ((data[1+6] == 0x85) & (data[2+6] == 0x03)): raise Exceptions.QuantityInvalidException("Value invalid"); if ((data[1+6] == 0x85) & (data[2+6] == 0x04)): raise Exceptions.ModbusException("error reading"); return True
Write single Coil to Master device (Function code 5) starting_address: Coil to be written value: Coil Value to be written
entailment
def write_multiple_coils(self, starting_address, values): """ Write multiple coils to Master device (Function code 15) starting_address : First coil to be written values: Coil Values [0..quantity-1] to be written """ self.__transactionIdentifier+=1 if (self.__ser is not None): if (self.__ser.closed): raise Exception.SerialPortNotOpenedException("serial port not opened"); function_code = 15 length = 6; transaction_identifier_lsb = self.__transactionIdentifier&0xFF transaction_identifier_msb = ((self.__transactionIdentifier&0xFF00) >> 8) length_lsb = length&0xFF length_msb = (length&0xFF00) >> 8 starting_address_lsb = starting_address&0xFF starting_address_msb = (starting_address&0xFF00) >> 8 quantityLSB = len(values)&0xFF quantityMSB = (len(values)&0xFF00) >> 8 valueToWrite = list() singleCoilValue = 0; for i in range(0, len(values)): if ((i % 8) == 0): if i > 0: valueToWrite.append(singleCoilValue) singleCoilValue = 0; if (values[i] == True): coilValue = 1 else: coilValue = 0 singleCoilValue = ((coilValue)<<(i%8) | (singleCoilValue)); valueToWrite.append(singleCoilValue) if (self.__ser is not None): data = bytearray([self.__unitIdentifier, function_code, starting_address_msb, starting_address_lsb, quantityMSB, quantityLSB]) data.append(len(valueToWrite)) #Bytecount for i in range (0, len(valueToWrite)): data.append(valueToWrite[i]&0xFF) crc = self.__calculateCRC(data, len(data), 0) crcLSB = crc&0xFF crcMSB = (crc&0xFF00) >> 8 data.append(crcLSB) data.append(crcMSB) self.__ser.write(data) bytes_to_read = 8 data = self.__ser.read(bytes_to_read) b=bytearray(data) data = b if (len(data) < bytes_to_read): raise Exceptions.TimeoutError('Read timeout Exception') if ((data[1] == 0x8F) & (data[2] == 0x01)): raise Exceptions.function_codeNotSupportedException("Function code not supported by master"); if ((data[1] == 0x8F) & (data[2] == 0x02)): raise Exceptions.starting_addressInvalidException("Starting address invalid or starting address + quantity invalid"); if ((data[1] == 0x8F) & (data[2] == 0x03)): raise Exceptions.QuantityInvalidException("quantity invalid"); if ((data[1] == 0x8F) & (data[2] == 0x04)): raise Exceptions.ModbusException("error reading"); crc = self.__calculateCRC(data, len(data) - 2, 0) crcLSB = crc&0xFF crcMSB = (crc&0xFF00) >> 8 if ((crcLSB != data[len(data)-2]) & (crcMSB != data[len(data)-1])): raise Exceptions.CRCCheckFailedException("CRC check failed"); if data[1] == self.__unitIdentifier: return True else: return False else: protocolIdentifierLSB = 0x00; protocolIdentifierMSB = 0x00; length_lsb = 0x06; length_msb = 0x00; data = bytearray([transaction_identifier_msb, transaction_identifier_lsb, protocolIdentifierMSB, protocolIdentifierLSB, length_msb, length_lsb, self.__unitIdentifier, function_code, starting_address_msb, starting_address_lsb, quantityMSB, quantityLSB]) data.append(len(valueToWrite)) #Bytecount for i in range (0, len(valueToWrite)): data.append(valueToWrite[i]&0xFF) self.__tcpClientSocket.send(data) bytes_to_read = 12 self.__receivedata = bytearray() try: while (len(self.__receivedata) == 0): pass except Exception: raise Exception('Read Timeout') data = bytearray(self.__receivedata) if ((data[1] == 0x8F) & (data[2] == 0x01)): raise Exceptions.function_codeNotSupportedException("Function code not supported by master"); if ((data[1] == 0x8F) & (data[2] == 0x02)): raise Exceptions.starting_addressInvalidException("Starting address invalid or starting address + quantity invalid"); if ((data[1] == 0x8F) & (data[2] == 0x03)): raise Exceptions.QuantityInvalidException("quantity invalid"); if ((data[1] == 0x8F) & (data[2] == 0x04)): raise Exceptions.ModbusException("error reading"); return True
Write multiple coils to Master device (Function code 15) starting_address : First coil to be written values: Coil Values [0..quantity-1] to be written
entailment
def services(self, service_group=None): """ Args: service_group: optional name of service group Returns: if service_group is omitted or None, flattened dict of all service records in the service registry if service_group is present, dict of service records in that group """ # Specific service group requested if service_group is not None: if service_group not in EFConfig.SERVICE_GROUPS: raise RuntimeError("service registry: {} doesn't have '{}' section listed in EFConfig".format( self._service_registry_file, service_group)) else: return self.service_registry_json[service_group] # Specific service group not requested - flatten and return all service records else: result = dict() for service_group in EFConfig.SERVICE_GROUPS: result.update(self.service_registry_json[service_group]) return result
Args: service_group: optional name of service group Returns: if service_group is omitted or None, flattened dict of all service records in the service registry if service_group is present, dict of service records in that group
entailment
def iter_services(self, service_group=None): """ Args: service_group: optional name of service group Returns: if service_group is omitted or None, an Iterator over all flattened service records in the service registry if service_group is present, an Iterator over all service records in that group """ if service_group is not None: if service_group not in EFConfig.SERVICE_GROUPS: raise RuntimeError("service registry: {} doesn't have '{}' section listed in EFConfig".format( self._service_registry_file, service_group)) return self.service_registry_json[service_group].iteritems() else: return self.services().iteritems()
Args: service_group: optional name of service group Returns: if service_group is omitted or None, an Iterator over all flattened service records in the service registry if service_group is present, an Iterator over all service records in that group
entailment
def valid_envs(self, service_name): """ Args: service_name: the name of the service in the service registry Returns: a list of strings - all the valid environments for 'service' Raises: RuntimeError if the service wasn't found """ service_record = self.service_record(service_name) if service_record is None: raise RuntimeError("service registry doesn't have service: {}".format(service_name)) # Return empty list if service has no "environments" section if not (service_record.has_key("environments")): return [] # Otherwise gather up the envs service_record_envs = service_record["environments"] result = [] for service_env in service_record_envs: if service_env not in EFConfig.PROTECTED_ENVS and service_env in EFConfig.EPHEMERAL_ENVS: result.extend((lambda env=service_env: [env + str(x) for x in range(EFConfig.EPHEMERAL_ENVS[env])])()) else: result.append(service_env) return result
Args: service_name: the name of the service in the service registry Returns: a list of strings - all the valid environments for 'service' Raises: RuntimeError if the service wasn't found
entailment
def service_record(self, service_name): """ Args: service_name: the name of the service in the service registry Returns: the entire service record from the service registry or None if the record was not found """ if not self.services().has_key(service_name): return None return self.services()[service_name]
Args: service_name: the name of the service in the service registry Returns: the entire service record from the service registry or None if the record was not found
entailment
def service_group(self, service_name): """ Args: service_name: the name of the service in the service registry Returns: the name of the group the service is in, or None of the service was not found """ for group in EFConfig.SERVICE_GROUPS: if self.services(group).has_key(service_name): return group return None
Args: service_name: the name of the service in the service registry Returns: the name of the group the service is in, or None of the service was not found
entailment
def service_region(self, service_name): """ Args: service_name: the name of the service in the service registry Returns: the region the service is in, or EFConfig.DEFAULT_REGION if the region was not found """ if not self.services()[service_name].has_key("region"): return EFConfig.DEFAULT_REGION else: return self.services()[service_name]["region"]
Args: service_name: the name of the service in the service registry Returns: the region the service is in, or EFConfig.DEFAULT_REGION if the region was not found
entailment
def easy_train_and_evaluate(hyper_params, Model=None, create_loss=None, training_data=None, validation_data=None, inline_plotting=False, session_config=None, log_suffix=None, continue_training=False, continue_with_specific_checkpointpath=None): """ Train and evaluate your model without any boilerplate code. 1) Write your data using the starttf.tfrecords.autorecords.write_data method. 2) Create your hyper parameter file containing all required fields and then load it using starttf.utils.hyper_params.load_params method. Minimal Sample Hyperparams File: {"train": { "learning_rate": { "type": "const", "start_value": 0.001 }, "optimizer": { "type": "adam" }, "batch_size": 1024, "iters": 10000, "summary_iters": 100, "checkpoint_path": "checkpoints/mnist", "tf_records_path": "data/.records/mnist" } } 3) Pass everything required to this method and that's it. :param hyper_params: The hyper parameters obejct loaded via starttf.utils.hyper_params.load_params :param Model: A keras model. :param create_loss: A create_loss function like that in starttf.examples.mnist.loss. :param inline_plotting: When you are using jupyter notebooks you can tell it to plot the loss directly inside the notebook. :param continue_training: Bool, continue last training in the checkpoint path specified in the hyper parameters. :param session_config: A configuration for the session. :param log_suffix: A suffix for the log folder, so you can remember what was special about the run. :return: """ time_stamp = datetime.datetime.fromtimestamp(time.time()).strftime('%Y-%m-%d_%H.%M.%S') chkpt_path = hyper_params.train.checkpoint_path + "/" + time_stamp if log_suffix is not None: chkpt_path = chkpt_path + "_" + log_suffix if session_config is None: session_config = get_default_config() if continue_with_specific_checkpointpath: chkpt_path = hyper_params.train.checkpoint_path + "/" + continue_with_specific_checkpointpath print("Continue with checkpoint: {}".format(chkpt_path)) elif continue_training: chkpts = sorted([name for name in os.listdir(hyper_params.train.checkpoint_path)]) chkpt_path = hyper_params.train.checkpoint_path + "/" + chkpts[-1] print("Latest found checkpoint: {}".format(chkpt_path)) if not os.path.exists(chkpt_path): os.makedirs(chkpt_path) # If hyperparam config is used, load and save code if Model is None: model_backup = os.path.join(chkpt_path, "model.py") copyfile(hyperparams["arch"]["model"].replace(".", os.sep), model_backup) arch_model = __import__(hyperparams["arch"]["model"], fromlist=["Model"]) Model = arch_model.Model if create_loss is None: loss_backup = os.path.join(chkpt_path, "loss.py") copyfile(hyperparams["arch"]["loss"].replace(".", os.sep), loss_backup) arch_loss = __import__(hyperparams["arch"]["loss"], fromlist=["create_loss"]) create_loss = arch_loss.create_loss # Load training data print("Load data") if training_data is None: training_data = create_input_fn(os.path.join(hyper_params.train.tf_records_path, PHASE_TRAIN), hyper_params.train.batch_size) if validation_data is None: validation_data = create_input_fn(os.path.join(hyper_params.train.tf_records_path, PHASE_VALIDATION), hyper_params.train.batch_size) # Write hyper parameters to be able to track what config you had. with open(chkpt_path + "/hyperparameters.json", "w") as json_file: json_file.write(json.dumps(hyper_params.to_dict(), indent=4, sort_keys=True)) estimator_spec = create_tf_estimator_spec(chkpt_path, Model, create_loss, inline_plotting) # Create a run configuration config = None if hyper_params.train.get("distributed", False): distribution = tf.contrib.distribute.MirroredStrategy() config = tf.estimator.RunConfig(model_dir=chkpt_path, save_summary_steps=hyper_params.train.summary_steps, train_distribute=distribution, save_checkpoints_steps=hyper_params.train.save_checkpoint_steps, keep_checkpoint_max=hyper_params.train.keep_checkpoint_max, keep_checkpoint_every_n_hours=1) else: config = tf.estimator.RunConfig(session_config=session_config, model_dir=chkpt_path, save_summary_steps=hyper_params.train.summary_steps, save_checkpoints_steps=hyper_params.train.save_checkpoint_steps, keep_checkpoint_max=hyper_params.train.keep_checkpoint_max, keep_checkpoint_every_n_hours=1) # Create the estimator. estimator = None if hyper_params.train.get("warm_start_checkpoint", None) is not None: warm_start_dir = hyper_params.train.warm_start_checkpoint estimator = tf.estimator.Estimator(estimator_spec, config=config, warm_start_from=warm_start_dir, params=hyper_params) else: estimator = tf.estimator.Estimator(estimator_spec, config=config, params=hyper_params) # Specify training and actually train. throttle_secs = hyper_params.train.get("throttle_secs", 120) train_spec = tf.estimator.TrainSpec(input_fn=training_data, max_steps=hyper_params.train.steps) eval_spec = tf.estimator.EvalSpec(input_fn=validation_data, throttle_secs=throttle_secs) print("Start training") tf.estimator.train_and_evaluate(estimator, train_spec, eval_spec) return estimator
Train and evaluate your model without any boilerplate code. 1) Write your data using the starttf.tfrecords.autorecords.write_data method. 2) Create your hyper parameter file containing all required fields and then load it using starttf.utils.hyper_params.load_params method. Minimal Sample Hyperparams File: {"train": { "learning_rate": { "type": "const", "start_value": 0.001 }, "optimizer": { "type": "adam" }, "batch_size": 1024, "iters": 10000, "summary_iters": 100, "checkpoint_path": "checkpoints/mnist", "tf_records_path": "data/.records/mnist" } } 3) Pass everything required to this method and that's it. :param hyper_params: The hyper parameters obejct loaded via starttf.utils.hyper_params.load_params :param Model: A keras model. :param create_loss: A create_loss function like that in starttf.examples.mnist.loss. :param inline_plotting: When you are using jupyter notebooks you can tell it to plot the loss directly inside the notebook. :param continue_training: Bool, continue last training in the checkpoint path specified in the hyper parameters. :param session_config: A configuration for the session. :param log_suffix: A suffix for the log folder, so you can remember what was special about the run. :return:
entailment
def create_prediction_estimator(hyper_params, model, checkpoint_path=None): """ Create an estimator for prediction purpose only. :param hyper_params: The hyper params file. :param model: The keras model. :param checkpoint_path: (Optional) Path to the specific checkpoint to use. :return: """ if checkpoint_path is None: chkpts = sorted([name for name in os.listdir(hyper_params.train.checkpoint_path)]) checkpoint_path = hyper_params.train.checkpoint_path + "/" + chkpts[-1] print("Latest found checkpoint: {}".format(checkpoint_path)) estimator_spec = create_tf_estimator_spec(checkpoint_path, model, create_loss=None) # Create the estimator. estimator = tf.estimator.Estimator(estimator_spec, model_dir=checkpoint_path, params=hyper_params) return estimator
Create an estimator for prediction purpose only. :param hyper_params: The hyper params file. :param model: The keras model. :param checkpoint_path: (Optional) Path to the specific checkpoint to use. :return:
entailment
def ef_plugin(service_name): """ Decorator for ef plugin classes. Any wrapped classes should contain a run() method which executes the plugin code. Args: service_name (str): The name of the service being extended. Example: @ef_plugin('ef-generate') class NewRelicPlugin(object): def run(self): exec_code() """ def class_rebuilder(cls): class EFPlugin(cls): """ Base class of ef-plugins. Defines which service is extended and provides access to the current instance of EFContext to the plugin. Args: context (obj:EFContext): Instance of EFContext created by ef-open command line tool clients (dict): Dictionary of boto3 clients created by ef_utils.create_aws_clients() """ def __init__(self, context, clients): self.service = service_name self.context = context self.clients = clients self.oInstance = cls() def __getattribute__(self, s): """ This is called whenever any attribute of a EFPlugin object is accessed. This function first tries to get the attribute off EFPlugin. If it fails then it tries to fetch the attribute from self.oInstance (an instance of the decorated class). """ try: x = super(EFPlugin, self).__getattribute__(s) except AttributeError: pass else: return x return self.oInstance.__getattribute__(s) return EFPlugin return class_rebuilder
Decorator for ef plugin classes. Any wrapped classes should contain a run() method which executes the plugin code. Args: service_name (str): The name of the service being extended. Example: @ef_plugin('ef-generate') class NewRelicPlugin(object): def run(self): exec_code()
entailment
def run_plugins(context_obj, boto3_clients): """ Executes all loaded plugins designated for the service calling the function. Args: context_obj (obj:EFContext): The EFContext object created by the service. boto3_clients (dict): Dictionary of boto3 clients created by ef_utils.create_aws_clients() """ def print_if_verbose(message): if context_obj.verbose: print(message) service_name = os.path.basename(sys.argv[0]).replace(".py", "") try: import plugins except ImportError: print_if_verbose("no plugins detected.") return else: for plugin_importer, plugin_name, plugin_ispkg in pkgutil.iter_modules(plugins.__path__): if plugin_ispkg: plugin_package = importlib.import_module("plugins.{}".format(plugin_name)) for importer, modname, ispkg in pkgutil.iter_modules(plugin_package.__path__): plugin_module = importlib.import_module("plugins.{}.{}".format(plugin_name, modname)) for name, obj in inspect.getmembers(plugin_module): if inspect.isclass(obj) and obj.__name__ == "EFPlugin": plugin_class = getattr(plugin_module, name) plugin_instance = plugin_class(context=context_obj, clients=boto3_clients) if plugin_instance.service == service_name: print_if_verbose("plugin '{}' loaded".format(plugin_name)) if not context_obj.commit: print_if_verbose("dryrun: skipping plugin execution.") else: try: plugin_instance.run() except AttributeError: print("error executing plugin '{}'".format(modname))
Executes all loaded plugins designated for the service calling the function. Args: context_obj (obj:EFContext): The EFContext object created by the service. boto3_clients (dict): Dictionary of boto3 clients created by ef_utils.create_aws_clients()
entailment
def easy_train_and_evaluate(hyper_params, Model=None, define_loss_fn=None, training_data=None, validation_data=None, continue_training=False, session_config=None, log_suffix=None, continue_with_specific_checkpointpath=None): """ Train and evaluate your model without any boilerplate code. 1) Write your data using the starttf.tfrecords.autorecords.write_data method. 2) Create your hyper parameter file containing all required fields and then load it using starttf.utils.hyper_params.load_params method. Minimal Sample Hyperparams File: {"train": { "learning_rate": { "type": "const", "start_value": 0.001 }, "optimizer": { "type": "adam" }, "batch_size": 1024, "iters": 10000, "summary_iters": 100, "checkpoint_path": "checkpoints/mnist", "tf_records_path": "data/.records/mnist" } } 3) Pass everything required to this method and that's it. :param hyper_params: The hyper parameters obejct loaded via starttf.utils.hyper_params.load_params :param Model: A keras model. :param create_loss: A create_loss function like that in starttf.examples.mnist.loss. :param inline_plotting: When you are using jupyter notebooks you can tell it to plot the loss directly inside the notebook. :param continue_training: Bool, continue last training in the checkpoint path specified in the hyper parameters. :param session_config: A configuration for the session. :param log_suffix: A suffix for the log folder, so you can remember what was special about the run. :return: """ time_stamp = datetime.datetime.fromtimestamp(time.time()).strftime('%Y-%m-%d_%H.%M.%S') chkpt_path = hyper_params.train.checkpoint_path + "/" + time_stamp if log_suffix is not None: chkpt_path = chkpt_path + "_" + log_suffix if session_config is None: session_config = get_default_config() if continue_with_specific_checkpointpath: chkpt_path = hyper_params.train.checkpoint_path + "/" + continue_with_specific_checkpointpath print("Continue with checkpoint: {}".format(chkpt_path)) elif continue_training: chkpts = sorted([name for name in os.listdir(hyper_params.train.checkpoint_path)]) chkpt_path = hyper_params.train.checkpoint_path + "/" + chkpts[-1] print("Latest found checkpoint: {}".format(chkpt_path)) if not os.path.exists(chkpt_path): os.makedirs(chkpt_path) # If hyperparam config is used if Model is None: arch_model = __import__(hyperparams.arch.model, fromlist=["Model"]) Model = arch_model.Model if define_loss_fn is None and hyperparams.arch.get("loss", None) is not None: arch_loss = __import__(hyperparams.arch.loss, fromlist=["define_loss_fn"]) define_loss_fn = arch_loss.define_loss_fn if training_data is None and hyperparams.problem.get("prepare", None) is not None: prepare = __import__(hyperparams.problem.prepare, fromlist=["Sequence"]) training_data = prepare.Sequence(hyperparams, PHASE_TRAIN) validation_data = prepare.Sequence(hyperparams, PHASE_VALIDATION) # TODO save code # Write hyper parameters to be able to track what config you had. with open(chkpt_path + "/hyperparameters.json", "w") as json_file: json_file.write(json.dumps(hyper_params.to_dict(), indent=4, sort_keys=True)) if training_data is not None: hyper_params.train.steps = hyper_params.train.epochs * len(training_data) losses = {} metrics = {} if define_loss_fn is None: losses = hyper_params.train.loss.to_dict() metrics = hyper_params.train.metrics.to_dict() else: losses, metrics = define_loss_fn(hyper_params) callbacks = create_keras_callbacks(hyper_params, chkpt_path) optimizer, lr_sheduler = create_keras_optimizer(hyper_params) callbacks.append(lr_sheduler) if training_data is None: train_features, train_labels = create_input_fn(os.path.join(hyper_params.train.tf_records_path, PHASE_TRAIN), hyper_params.train.batch_size)().make_one_shot_iterator().get_next() validation_data = create_input_fn(os.path.join(hyper_params.train.tf_records_path, PHASE_VALIDATION), hyper_params.train.batch_size)().make_one_shot_iterator().get_next() model = Model(hyper_params) input_tensor = {k: tf.keras.layers.Input(shape=train_features[k].get_shape().as_list(), name=k) for k in train_features} target_placeholders = {k: tf.placeholder(shape=(None,) + train_labels[k].shape[1:], dtype=train_labels[k].dtype, name=k + "_placeholder") for k in train_labels} model = model.create_keras_model(input_tensor, training=True) # model.metrics_names = [k for k in metrics] model.compile(loss=losses, optimizer=optimizer, metrics=[rename_fn(v, name=k) for k, v in metrics.iteritems()], target_tensors=target_placeholders) tf.keras.backend.get_session().run(tf.global_variables_initializer()) model.fit(train_features, train_labels, validation_data=validation_data, batch_size=hyper_params.train.batch_size, steps_per_epoch=hyper_params.train.get("steps_per_epoch", 1), epochs=hyper_params.train.get("epochs", 50), validation_steps=hyper_params.train.get("validation_steps", 1), callbacks=callbacks, verbose=1) else: # first batches features #features = training_data[0][0] #model._set_inputs({k: tf.zeros(features[k].shape) for k in features}) model = Model(hyper_params) train_features = training_data[0][0] train_labels = training_data[0][1] input_tensor = {k: tf.keras.layers.Input(shape=train_features[k].shape[1:], name=k) for k in train_features} target_placeholders = {k: tf.placeholder(shape=(None,) + train_labels[k].shape[1:], dtype=train_labels[k].dtype, name=k + "_placeholder") for k in train_labels} model = model.create_keras_model(input_tensor, training=True) # model.metrics_names = [k for k in metrics] model.compile(loss=losses, optimizer=optimizer, metrics=[metrics[k] for k in metrics], target_tensors=target_placeholders) tf.keras.backend.get_session().run(tf.global_variables_initializer()) model.fit_generator(training_data, validation_data=validation_data, epochs=hyper_params.train.get("epochs", 50), callbacks=callbacks, workers=2, use_multiprocessing=False, shuffle=True, verbose=1) return chkpt_path
Train and evaluate your model without any boilerplate code. 1) Write your data using the starttf.tfrecords.autorecords.write_data method. 2) Create your hyper parameter file containing all required fields and then load it using starttf.utils.hyper_params.load_params method. Minimal Sample Hyperparams File: {"train": { "learning_rate": { "type": "const", "start_value": 0.001 }, "optimizer": { "type": "adam" }, "batch_size": 1024, "iters": 10000, "summary_iters": 100, "checkpoint_path": "checkpoints/mnist", "tf_records_path": "data/.records/mnist" } } 3) Pass everything required to this method and that's it. :param hyper_params: The hyper parameters obejct loaded via starttf.utils.hyper_params.load_params :param Model: A keras model. :param create_loss: A create_loss function like that in starttf.examples.mnist.loss. :param inline_plotting: When you are using jupyter notebooks you can tell it to plot the loss directly inside the notebook. :param continue_training: Bool, continue last training in the checkpoint path specified in the hyper parameters. :param session_config: A configuration for the session. :param log_suffix: A suffix for the log folder, so you can remember what was special about the run. :return:
entailment
def crop(img, start_y, start_x, h, w): """ Crop an image given the top left corner. :param img: The image :param start_y: The top left corner y coord :param start_x: The top left corner x coord :param h: The result height :param w: The result width :return: The cropped image. """ return img[start_y:start_y + h, start_x:start_x + w, :].copy()
Crop an image given the top left corner. :param img: The image :param start_y: The top left corner y coord :param start_x: The top left corner x coord :param h: The result height :param w: The result width :return: The cropped image.
entailment
def resize_image_with_crop_or_pad(img, target_height, target_width): """ Crops and/or pads an image to a target width and height. Resizes an image to a target width and height by either cropping the image or padding it with zeros. NO CENTER CROP. NO CENTER PAD. (Just fill bottom right or crop bottom right) :param img: Numpy array representing the image. :param target_height: Target height. :param target_width: Target width. :return: The cropped and padded image. """ h, w = target_height, target_width max_h, max_w, c = img.shape # crop img = crop_center(img, min(max_h, h), min(max_w, w)) # pad padded_img = np.zeros(shape=(h, w, c), dtype=img.dtype) padded_img[:img.shape[0], :img.shape[1], :img.shape[2]] = img return padded_img
Crops and/or pads an image to a target width and height. Resizes an image to a target width and height by either cropping the image or padding it with zeros. NO CENTER CROP. NO CENTER PAD. (Just fill bottom right or crop bottom right) :param img: Numpy array representing the image. :param target_height: Target height. :param target_width: Target width. :return: The cropped and padded image.
entailment
def _rotatedRectWithMaxArea(w, h, angle): """ Given a rectangle of size wxh that has been rotated by 'angle' (in radians), computes the width and height of the largest possible axis-aligned rectangle (maximal area) within the rotated rectangle. Answer from: https://stackoverflow.com/questions/16702966/rotate-image-and-crop-out-black-borders """ if w <= 0 or h <= 0: return 0,0 width_is_longer = w >= h side_long, side_short = (w,h) if width_is_longer else (h,w) # since the solutions for angle, -angle and 180-angle are all the same, # if suffices to look at the first quadrant and the absolute values of sin,cos: sin_a, cos_a = abs(math.sin(angle)), abs(math.cos(angle)) if side_short <= 2.*sin_a*cos_a*side_long or abs(sin_a-cos_a) < 1e-10: # half constrained case: two crop corners touch the longer side, # the other two corners are on the mid-line parallel to the longer line x = 0.5*side_short wr,hr = (x/sin_a,x/cos_a) if width_is_longer else (x/cos_a,x/sin_a) else: # fully constrained case: crop touches all 4 sides cos_2a = cos_a*cos_a - sin_a*sin_a wr,hr = (w*cos_a - h*sin_a)/cos_2a, (h*cos_a - w*sin_a)/cos_2a return wr,hr
Given a rectangle of size wxh that has been rotated by 'angle' (in radians), computes the width and height of the largest possible axis-aligned rectangle (maximal area) within the rotated rectangle. Answer from: https://stackoverflow.com/questions/16702966/rotate-image-and-crop-out-black-borders
entailment
def rotate_img_and_crop(img, angle): """ Rotate an image and then crop it so that there is no black area. :param img: The image to rotate. :param angle: The rotation angle in degrees. :return: The rotated and cropped image. """ h, w, _ = img.shape img = scipy.ndimage.interpolation.rotate(img, angle) w, h = _rotatedRectWithMaxArea(w, h, math.radians(angle)) return crop_center(img, int(h), int(w))
Rotate an image and then crop it so that there is no black area. :param img: The image to rotate. :param angle: The rotation angle in degrees. :return: The rotated and cropped image.
entailment
def diff_string_templates(string_a, string_b): """ Determine the diff of two strings. Return an empty string if the strings are identical, and the diff output string if they are not. """ s1 = string_a.strip().splitlines() s2 = string_b.strip().splitlines() diffs = unified_diff(s2, s1, fromfile='deployed', tofile='local', lineterm='') return '\n'.join(diffs)
Determine the diff of two strings. Return an empty string if the strings are identical, and the diff output string if they are not.
entailment
def render_local_template(service_name, environment, repo_root, template_file): """ Render a given service's template for a given environment and return it """ cmd = 'cd {} && ef-cf {} {} --devel --verbose'.format(repo_root, template_file, environment) p = subprocess.Popen(cmd, shell=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE) stdout, stderr = p.communicate() if p.returncode != 0: stderr = indentify('\n{}'.format(stderr)) stdout = indentify('\n{}'.format(stdout)) raise Exception('Service: `{}`, Env: `{}`, Msg: `{}{}`' .format(service_name, environment, stderr, stdout)) logger.debug('Rendered template for `%s` in `%s`', template_file, environment) r = re.match(r".*(^{.*^})$", stdout, re.MULTILINE | re.DOTALL) return jsonify(json.loads(r.group(1)))
Render a given service's template for a given environment and return it
entailment
def fetch_current_cloudformation_template(service_name, environment, cf_client): """ Fetch the currently-deployed template for the given service in the given environment and return it. """ stack_name = get_stack_name(environment, service_name) logger.debug('Fetching template for `%s`', stack_name) result = cf_client.get_template(StackName=stack_name) return jsonify(result['TemplateBody'])
Fetch the currently-deployed template for the given service in the given environment and return it.
entailment
def diff_sevice_by_text(service_name, service, environment, cf_client, repo_root): """ Render the local template and compare it to the template that was last applied in the target environment. """ global ret_code logger.info('Investigating textual diff for `%s`:`%s` in environment `%s`', service['type'], service_name, environment) try: local_template = render_local_template(service_name, environment, repo_root, service['template_file']) current_template = fetch_current_cloudformation_template( service_name, environment, cf_client) except Exception as e: ret_code = 2 logger.error(e) return ret = diff_string_templates(local_template, current_template) if not ret: logger.info('Deployed service `%s` in environment `%s` matches ' 'the local template.', service_name, environment) else: ret_code = 1 logger.error('Service `%s` in environment `%s` differs from ' 'the local template.', service_name, environment) logger.info('Change details:\n %s', indentify(ret))
Render the local template and compare it to the template that was last applied in the target environment.
entailment
def diff_sevice_by_changeset(service_name, service, environment, cf_client, repo_root): """ If an ef-cf call fails, the error will be logged, the retcode set to 2, but the function will run to completion and return the list of non-error results. """ global ret_code logger.info('Investigating changeset for `%s`:`%s` in environment `%s`', service['type'], service_name, environment) delete_any_existing_changesets(cf_client, service_name, environment) try: changeset = generate_changeset(service_name, environment, repo_root, service['template_file']) except Exception as e: ret_code = 2 logger.error(e) return wait_for_changeset_creation(cf_client, changeset['Id'], changeset['StackId']) logger.info('Created Changeset ID: `%s`', changeset['Id']) desc = cf_client.describe_change_set( ChangeSetName=changeset['Id'], StackName=changeset['StackId']) cf_client.delete_change_set( ChangeSetName=changeset['Id'], StackName=changeset['StackId']) if changeset_is_empty(desc): logger.info('Deployed service `%s` in environment `%s` matches ' 'the local template.', service_name, environment) else: ret_code = 1 logger.error('Service `%s` in environment `%s` differs from ' 'the local template.', service_name, environment) details = jsonify(desc['Changes']) logger.info('Change details:\n %s', indentify(details))
If an ef-cf call fails, the error will be logged, the retcode set to 2, but the function will run to completion and return the list of non-error results.
entailment
def get_cloudformation_client(service_name, environment_name): """ Given a service name and an environment name, return a boto CloudFormation client object. """ region = service_registry.service_region(service_name) if whereami() == 'ec2': profile = None else: profile = get_account_alias(environment_name) clients = create_aws_clients(region, profile, 'cloudformation') return clients['cloudformation']
Given a service name and an environment name, return a boto CloudFormation client object.
entailment
def evaluate_service_changes(services, envs, repo_root, func): """ Given a dict of services, and a list of environments, apply the diff function to evaluate the differences between the target environments and the rendered templates. Sub-services (names with '.' in them) are skipped. """ for service_name, service in services.iteritems(): for env_category in service['environments']: if env_category not in get_env_categories(envs): logger.debug('Skipping not-included environment `%s` for service `%s`', env_category, service_name) continue environment = generate_test_environment_name(env_category) cf_client = get_cloudformation_client(service_name, environment) func(service_name, service, environment, cf_client, repo_root)
Given a dict of services, and a list of environments, apply the diff function to evaluate the differences between the target environments and the rendered templates. Sub-services (names with '.' in them) are skipped.
entailment
def get_matching_service_template_file(service_name, template_files): """ Return the template file that goes with the given service name, or return None if there's no match. Subservices return the parent service's file. """ # If this is a subservice, use the parent service's template service_name = service_name.split('.')[0] if service_name in template_files: return template_files[service_name] return None
Return the template file that goes with the given service name, or return None if there's no match. Subservices return the parent service's file.
entailment
def get_dict_registry_services(registry, template_files, warn_missing_files=True): """ Return a dict mapping service name to a dict containing the service's type ('fixtures', 'platform_services', 'application_services', 'internal_services'), the template file's absolute path, and a list of environments to which the service is intended to deploy. Service names that appear twice in the output list will emit a warning and ignore the latter records. Services which have no template file will not appear in the returned dict. If the `warn_missing_files` boolean is True these files will emit a warning. """ with open(registry) as fr: parsed_registry = json.load(fr) services = {} for type, type_services in parsed_registry.iteritems(): for name, service in type_services.iteritems(): if name in services: logger.warning("Template name appears twice, ignoring later items: `%s`", name) continue template_file = get_matching_service_template_file(name, template_files) if not template_file: if warn_missing_files: logger.warning("No template file for `%s` (%s) `%s`", type, service['type'], name) continue services[name] = { 'type': type, 'template_file': template_file, 'environments': service['environments'] } return services
Return a dict mapping service name to a dict containing the service's type ('fixtures', 'platform_services', 'application_services', 'internal_services'), the template file's absolute path, and a list of environments to which the service is intended to deploy. Service names that appear twice in the output list will emit a warning and ignore the latter records. Services which have no template file will not appear in the returned dict. If the `warn_missing_files` boolean is True these files will emit a warning.
entailment
def scan_dir_for_template_files(search_dir): """ Return a map of "likely service/template name" to "template file". This includes all the template files in fixtures and in services. """ template_files = {} cf_dir = os.path.join(search_dir, 'cloudformation') for type in os.listdir(cf_dir): template_dir = os.path.join(cf_dir, type, 'templates') for x in os.listdir(template_dir): name = os.path.splitext(x)[0] template_files[name] = os.path.join(template_dir, x) return template_files
Return a map of "likely service/template name" to "template file". This includes all the template files in fixtures and in services.
entailment
def generate_secret(length=32): """ Generate a random secret consisting of mixed-case letters and numbers Args: length (int): Length of the generated password Returns: a randomly generated secret string Raises: None """ alphabet = string.ascii_letters + string.digits random_bytes = os.urandom(length) indices = [int(len(alphabet) * (ord(byte) / 256.0)) for byte in random_bytes] return "".join([alphabet[index] for index in indices])
Generate a random secret consisting of mixed-case letters and numbers Args: length (int): Length of the generated password Returns: a randomly generated secret string Raises: None
entailment
def generate_secret_file(file_path, pattern, service, environment, clients): """ Generate a parameter files with it's secrets encrypted in KMS Args: file_path (string): Path to the parameter file to be encrypted pattern (string): Pattern to do fuzzy string matching service (string): Service to use KMS key to encrypt file environment (string): Environment to encrypt values clients (dict): KMS AWS client that has been instantiated Returns: None Raises: IOError: If the file does not exist """ changed = False with open(file_path) as json_file: data = json.load(json_file, object_pairs_hook=OrderedDict) try: for key, value in data["params"][environment].items(): if pattern in key: if "aws:kms:decrypt" in value: print("Found match, key {} but value is encrypted already; skipping...".format(key)) else: print("Found match, encrypting key {}".format(key)) encrypted_password = ef_utils.kms_encrypt(clients['kms'], service, environment, value) data["params"][environment][key] = format_secret(encrypted_password) changed = True except KeyError: ef_utils.fail("Error env: {} does not exist in parameters file".format(environment)) if changed: with open(file_path, "w") as encrypted_file: json.dump(data, encrypted_file, indent=2, separators=(',', ': ')) # Writing new line here so it conforms to WG14 N1256 5.1.1.1 (so github doesn't complain) encrypted_file.write("\n")
Generate a parameter files with it's secrets encrypted in KMS Args: file_path (string): Path to the parameter file to be encrypted pattern (string): Pattern to do fuzzy string matching service (string): Service to use KMS key to encrypt file environment (string): Environment to encrypt values clients (dict): KMS AWS client that has been instantiated Returns: None Raises: IOError: If the file does not exist
entailment
def handle_args_and_set_context(args): """ Args: args: the command line args, probably passed from main() as sys.argv[1:] Returns: a populated EFPWContext object Raises: RuntimeError: if repo or branch isn't as spec'd in ef_config.EF_REPO and ef_config.EF_REPO_BRANCH ValueError: if a parameter is invalid """ parser = argparse.ArgumentParser() parser.add_argument("service", help="name of service password is being generated for") parser.add_argument("env", help=", ".join(EFConfig.ENV_LIST)) group = parser.add_mutually_exclusive_group() group.add_argument("--decrypt", help="encrypted string to be decrypted", default="") group.add_argument("--plaintext", help="secret to be encrypted rather than a randomly generated one", default="") group.add_argument("--secret_file", help="json file containing secrets to be encrypted", default="") parser.add_argument("--match", help="used in conjunction with --secret_file to match against keys to be encrypted", default="") parser.add_argument("--length", help="length of generated password (default 32)", default=32) parsed_args = vars(parser.parse_args(args)) context = EFPWContext() try: context.env = parsed_args["env"] except ValueError as e: ef_utils.fail("Error in env: {}".format(e.message)) context.service = parsed_args["service"] context.decrypt = parsed_args["decrypt"] context.length = parsed_args["length"] context.plaintext = parsed_args["plaintext"] context.secret_file = parsed_args["secret_file"] context.match = parsed_args["match"] if context.match or context.secret_file: if not context.match or not context.secret_file: raise ValueError("Must have both --match and --secret_file flag") return context
Args: args: the command line args, probably passed from main() as sys.argv[1:] Returns: a populated EFPWContext object Raises: RuntimeError: if repo or branch isn't as spec'd in ef_config.EF_REPO and ef_config.EF_REPO_BRANCH ValueError: if a parameter is invalid
entailment
def tile_2d(input, k_x, k_y, name, reorder_required=True): """ A tiling layer like introduced in overfeat and huval papers. :param input: Your input tensor. :param k_x: The tiling factor in x direction. :param k_y: The tiling factor in y direction. :param name: The name of the layer. :param reorder_required: To implement an exact huval tiling you need reordering. However not using it is more efficient and when training from scratch setting this to false is highly recommended. :return: The output tensor. """ size = input.get_shape().as_list() c, h, w = size[3], size[1], size[2] batch_size = size[0] if batch_size is None: batch_size = -1 # Check if tiling is possible and define output shape. assert c % (k_x * k_y) == 0 tmp = input if reorder_required: output_channels = int(c / (k_x * k_y)) channels = tf.unstack(tmp, axis=-1) reordered_channels = [None for _ in range(len(channels))] for o in range(output_channels): for i in range(k_x * k_y): target = o + i * output_channels source = o * (k_x * k_y) + i reordered_channels[target] = channels[source] tmp = tf.stack(reordered_channels, axis=-1) # Actual tilining with tf.variable_scope(name) as scope: tmp = tf.transpose(tmp, [0, 2, 1, 3]) tmp = tf.reshape(tmp, (batch_size, w, int(h * k_y), int(c / (k_y)))) tmp = tf.transpose(tmp, [0, 2, 1, 3]) tmp = tf.reshape(tmp, (batch_size, int(h * k_y), int(w * k_x), int(c / (k_y * k_x)))) return tmp
A tiling layer like introduced in overfeat and huval papers. :param input: Your input tensor. :param k_x: The tiling factor in x direction. :param k_y: The tiling factor in y direction. :param name: The name of the layer. :param reorder_required: To implement an exact huval tiling you need reordering. However not using it is more efficient and when training from scratch setting this to false is highly recommended. :return: The output tensor.
entailment
def inverse_tile_2d(input, k_x, k_y, name): """ An inverse tiling layer. An inverse to the tiling layer can be of great use, since you can keep the resolution of your output low, but harness the benefits of the resolution of a higher level feature layer. If you insist on a source you can call it very lightly inspired by yolo9000 "passthrough layer". :param input: Your input tensor. (Assert input.shape[1] % k_y = 0 and input.shape[2] % k_x = 0) :param k_x: The tiling factor in x direction [int]. :param k_y: The tiling factor in y direction [int]. :param name: The name of the layer. :return: The output tensor of shape [batch_size, inp.height / k_y, inp.width / k_x, inp.channels * k_x * k_y]. """ batch_size, h, w, c = input.get_shape().as_list() if batch_size is None: batch_size = -1 # Check if tiling is possible and define output shape. assert w % k_x == 0 and h % k_y == 0 # Actual inverse tilining with tf.variable_scope(name) as scope: tmp = input tmp = tf.reshape(tmp, (batch_size, int(h * k_y), w, int(c * k_x))) tmp = tf.transpose(tmp, [0, 2, 1, 3]) tmp = tf.reshape(tmp, (batch_size, w, h, int(c * k_y * k_x))) tmp = tf.transpose(tmp, [0, 2, 1, 3]) return tmp
An inverse tiling layer. An inverse to the tiling layer can be of great use, since you can keep the resolution of your output low, but harness the benefits of the resolution of a higher level feature layer. If you insist on a source you can call it very lightly inspired by yolo9000 "passthrough layer". :param input: Your input tensor. (Assert input.shape[1] % k_y = 0 and input.shape[2] % k_x = 0) :param k_x: The tiling factor in x direction [int]. :param k_y: The tiling factor in y direction [int]. :param name: The name of the layer. :return: The output tensor of shape [batch_size, inp.height / k_y, inp.width / k_x, inp.channels * k_x * k_y].
entailment
def feature_passthrough(early_feat, late_feat, filters, name, kernel_size=(1, 1)): """ A feature passthrough layer inspired by yolo9000 and the inverse tiling layer. It can be proven, that this layer does the same as conv(concat(inverse_tile(early_feat), late_feat)). This layer has no activation function. :param early_feat: The early feature layer of shape [batch_size, h * s_x, w * s_y, _]. s_x and s_y are integers computed internally describing the scale between the layers. :param late_feat: The late feature layer of shape [batch_size, h, w, _]. :param filters: The number of convolution filters. :param name: The name of the layer. :param kernel_size: The size of the kernel. Default (1x1). :return: The output tensor of shape [batch_size, h, w, outputs] """ _, h_early, w_early, c_early = early_feat.get_shape().as_list() _, h_late, w_late, c_late = late_feat.get_shape().as_list() s_x = int(w_early / w_late) s_y = int(h_early / h_late) assert h_late * s_y == h_early and w_late * s_x == w_early with tf.variable_scope(name) as scope: early_conv = tf.layers.conv2d(early_feat, filters=filters, kernel_size=(s_x * kernel_size[0], s_y * kernel_size[1]), strides=(s_x, s_y), padding="same") late_conv = tf.layers.conv2d(late_feat, filters=filters, kernel_size=kernel_size, strides=(1, 1), padding="same") return early_conv + late_conv
A feature passthrough layer inspired by yolo9000 and the inverse tiling layer. It can be proven, that this layer does the same as conv(concat(inverse_tile(early_feat), late_feat)). This layer has no activation function. :param early_feat: The early feature layer of shape [batch_size, h * s_x, w * s_y, _]. s_x and s_y are integers computed internally describing the scale between the layers. :param late_feat: The late feature layer of shape [batch_size, h, w, _]. :param filters: The number of convolution filters. :param name: The name of the layer. :param kernel_size: The size of the kernel. Default (1x1). :return: The output tensor of shape [batch_size, h, w, outputs]
entailment
def upsampling_feature_passthrough(early_feat, late_feat, filters, name, kernel_size=(1, 1)): """ An upsampling feature passthrough layer inspired by yolo9000 and the tiling layer. It can be proven, that this layer does the same as conv(concat(early_feat, tile_2d(late_feat))). This layer has no activation function. :param early_feat: The early feature layer of shape [batch_size, h * s_x, w * s_y, _]. s_x and s_y are integers computed internally describing the scale between the layers. :param late_feat: The late feature layer of shape [batch_size, h, w, _]. :param filters: The number of convolution filters. :param name: The name of the layer. :param kernel_size: The size of the kernel. Default (1x1). :return: The output tensor of shape [batch_size, h * s_x, w * s_y, outputs] """ _, h_early, w_early, c_early = early_feat.get_shape().as_list() _, h_late, w_late, c_late = late_feat.get_shape().as_list() s_x = int(w_early / w_late) s_y = int(h_early / h_late) assert h_late * s_y == h_early and w_late * s_x == w_early with tf.variable_scope(name) as scope: tiled = tile_2d(late_feat, s_x, s_y, "tile_2d", reorder_required=False) concated = tf.concat([early_feat, tiled], axis=-1) return tf.layers.conv2d(concated, filters=filters, kernel_size=kernel_size, strides=(1, 1), padding="same")
An upsampling feature passthrough layer inspired by yolo9000 and the tiling layer. It can be proven, that this layer does the same as conv(concat(early_feat, tile_2d(late_feat))). This layer has no activation function. :param early_feat: The early feature layer of shape [batch_size, h * s_x, w * s_y, _]. s_x and s_y are integers computed internally describing the scale between the layers. :param late_feat: The late feature layer of shape [batch_size, h, w, _]. :param filters: The number of convolution filters. :param name: The name of the layer. :param kernel_size: The size of the kernel. Default (1x1). :return: The output tensor of shape [batch_size, h * s_x, w * s_y, outputs]
entailment
def load(self): """Loads the config""" try: with open(self._ef_site_config, 'r') as yml_file: return yaml.safe_load(yml_file) except (IOError, yaml.parser.ParserError) as error: print("Error: {}".format(error), file=sys.stderr) sys.exit(1)
Loads the config
entailment
def interpolate_loss(labels, loss1, loss2, interpolation_values): """ Interpolate two losses linearly. :param labels: A float tensor of shape [batch_size, ..., num_classes] representing the label class probabilities. :param loss1: A float tensor of shape [batch_size, ...] representing the loss1 for interpolation. :param loss2: A float tensor of shape [batch_size, ...] representing the loss2 for interpolation. :param interpolation_values: The values for each class how much focal loss should be interpolated in. :return: A tensor representing the weighted cross entropy. """ with tf.variable_scope("interpolate_focus_loss"): # Select the probs or weights with the labels. t = tf.reduce_sum(labels * interpolation_values, axis=-1) return (1 - t) * loss1 + t * loss2
Interpolate two losses linearly. :param labels: A float tensor of shape [batch_size, ..., num_classes] representing the label class probabilities. :param loss1: A float tensor of shape [batch_size, ...] representing the loss1 for interpolation. :param loss2: A float tensor of shape [batch_size, ...] representing the loss2 for interpolation. :param interpolation_values: The values for each class how much focal loss should be interpolated in. :return: A tensor representing the weighted cross entropy.
entailment
def alpha_balance_loss(labels, loss, alpha_weights): """ Calculate the alpha balanced cross_entropy. This means for each sample the cross entropy is calculated and then weighted by the class specific weight. :param labels: A float tensor of shape [batch_size, ..., num_classes] representing the label class probabilities. :param loss: A float tensor of shape [batch_size, ...] representing the loss that should be focused. :param alpha_weights: A float tensor of shape [1, ..., num_classes] (... is filled with ones to match number of dimensions to labels tensor) representing the weights for each class. :return: A tensor representing the weighted cross entropy. """ with tf.variable_scope("alpha_balance"): # Broadcast multiply labels with alpha weights to select weights and then reduce them along last axis. weights = tf.reduce_sum(labels * alpha_weights, axis=-1) return weights * loss
Calculate the alpha balanced cross_entropy. This means for each sample the cross entropy is calculated and then weighted by the class specific weight. :param labels: A float tensor of shape [batch_size, ..., num_classes] representing the label class probabilities. :param loss: A float tensor of shape [batch_size, ...] representing the loss that should be focused. :param alpha_weights: A float tensor of shape [1, ..., num_classes] (... is filled with ones to match number of dimensions to labels tensor) representing the weights for each class. :return: A tensor representing the weighted cross entropy.
entailment
def batch_alpha_balance_loss(labels, loss): """ Calculate the alpha balanced cross_entropy. This means for each sample the cross entropy is calculated and then weighted by the class specific weight. There is yet no paper for this type of loss. :param labels: A float tensor of shape [batch_size, ..., num_classes] representing the label class probabilities. :param loss: A float tensor of shape [batch_size, ...] representing the loss that should be focused. :return: A tensor representing the weighted cross entropy. """ with tf.variable_scope("batch_alpha_balance"): # Compute the occurrence probability for each class mu, _ = tf.nn.moments(labels, [0, 1, 2]) # For weighting a class should be down weighted by its occurrence probability. not_mu = 1 - mu # Select the class specific not_mu not_mu_class = tf.reduce_sum(labels * not_mu, axis=-1) return not_mu_class * loss
Calculate the alpha balanced cross_entropy. This means for each sample the cross entropy is calculated and then weighted by the class specific weight. There is yet no paper for this type of loss. :param labels: A float tensor of shape [batch_size, ..., num_classes] representing the label class probabilities. :param loss: A float tensor of shape [batch_size, ...] representing the loss that should be focused. :return: A tensor representing the weighted cross entropy.
entailment
def mask_loss(input_tensor, binary_tensor): """ Mask a loss by using a tensor filled with 0 or 1. :param input_tensor: A float tensor of shape [batch_size, ...] representing the loss/cross_entropy :param binary_tensor: A float tensor of shape [batch_size, ...] representing the mask. :return: A float tensor of shape [batch_size, ...] representing the masked loss. """ with tf.variable_scope("mask_loss"): mask = tf.cast(tf.cast(binary_tensor, tf.bool), tf.float32) return input_tensor * mask
Mask a loss by using a tensor filled with 0 or 1. :param input_tensor: A float tensor of shape [batch_size, ...] representing the loss/cross_entropy :param binary_tensor: A float tensor of shape [batch_size, ...] representing the mask. :return: A float tensor of shape [batch_size, ...] representing the masked loss.
entailment
def mean_on_masked(loss, mask, epsilon=1e-8, axis=None): """ Average a loss correctly when it was masked. :param loss: A float tensor of shape [batch_size, ...] representing the (already masked) loss to be averaged. :param mask: A float tensor of shape [batch_size, ...] representing the mask. :param epsilon: Offset of log for numerical stability. :param axis: The dimensions to reduce. If None (the default), reduces all dimensions. Must be in the range [-rank(input_tensor), rank(input_tensor)). """ mask = tf.cast(tf.cast(mask, tf.bool), tf.float32) active_pixels = tf.reduce_sum(mask) active_pixels = tf_if(tf.equal(active_pixels, 0), epsilon, active_pixels) return tf.reduce_sum(loss, axis=axis) / active_pixels
Average a loss correctly when it was masked. :param loss: A float tensor of shape [batch_size, ...] representing the (already masked) loss to be averaged. :param mask: A float tensor of shape [batch_size, ...] representing the mask. :param epsilon: Offset of log for numerical stability. :param axis: The dimensions to reduce. If None (the default), reduces all dimensions. Must be in the range [-rank(input_tensor), rank(input_tensor)).
entailment
def mask_and_mean_loss(input_tensor, binary_tensor, axis=None): """ Mask a loss by using a tensor filled with 0 or 1 and average correctly. :param input_tensor: A float tensor of shape [batch_size, ...] representing the loss/cross_entropy :param binary_tensor: A float tensor of shape [batch_size, ...] representing the mask. :return: A float tensor of shape [batch_size, ...] representing the masked loss. :param axis: The dimensions to reduce. If None (the default), reduces all dimensions. Must be in the range [-rank(input_tensor), rank(input_tensor)). """ return mean_on_masked(mask_loss(input_tensor, binary_tensor), binary_tensor, axis=axis)
Mask a loss by using a tensor filled with 0 or 1 and average correctly. :param input_tensor: A float tensor of shape [batch_size, ...] representing the loss/cross_entropy :param binary_tensor: A float tensor of shape [batch_size, ...] representing the mask. :return: A float tensor of shape [batch_size, ...] representing the masked loss. :param axis: The dimensions to reduce. If None (the default), reduces all dimensions. Must be in the range [-rank(input_tensor), rank(input_tensor)).
entailment
def variance_corrected_loss(loss, sigma_2=None): """ Create a variance corrected loss. When summing variance corrected losses you get the same as multiloss. This is especially usefull for keras where when having multiple losses they are summed by keras. This multi-loss implementation is inspired by the Paper "Multi-Task Learning Using Uncertainty to Weight Losses for Scene Geometry and Semantics" by Kendall, Gal and Cipolla. :param loss: The loss that should be variance corrected. :param sigma_2: Optional a variance (sigma squared) to use. If none is provided it is learned. :return: The variance corrected loss. """ with tf.variable_scope("variance_corrected_loss"): sigma_cost = 0 if sigma_2 is None: # FIXME the paper has been updated Apr 2018, check if implementation is still valid. sigma = tf.get_variable(name="sigma", dtype=tf.float32, initializer=tf.constant(1.0), trainable=True) sigma_2 = tf.pow(sigma, 2) tf.summary.scalar("sigma2", sigma_2) sigma_cost = tf.log(sigma_2 + 1.0) return 0.5 / sigma_2 * loss + sigma_cost
Create a variance corrected loss. When summing variance corrected losses you get the same as multiloss. This is especially usefull for keras where when having multiple losses they are summed by keras. This multi-loss implementation is inspired by the Paper "Multi-Task Learning Using Uncertainty to Weight Losses for Scene Geometry and Semantics" by Kendall, Gal and Cipolla. :param loss: The loss that should be variance corrected. :param sigma_2: Optional a variance (sigma squared) to use. If none is provided it is learned. :return: The variance corrected loss.
entailment
def multiloss(losses, logging_namespace="multiloss", exclude_from_weighting=[]): """ Create a loss from multiple losses my mixing them. This multi-loss implementation is inspired by the Paper "Multi-Task Learning Using Uncertainty to Weight Losses for Scene Geometry and Semantics" by Kendall, Gal and Cipolla. :param losses: A dict containing all losses that should be merged. :param logging_namespace: Variable scope in which multiloss lives. :param exclude_from_weighting: A list of losses that are already weighted and should not be sigma weighted. :return: A single loss. """ with tf.variable_scope(logging_namespace): sum_loss = 0 for loss_name, loss in losses.items(): if loss_name not in exclude_from_weighting: with tf.variable_scope(loss_name) as scope: sum_loss += variance_corrected_loss(loss) else: sum_loss += loss return sum_loss
Create a loss from multiple losses my mixing them. This multi-loss implementation is inspired by the Paper "Multi-Task Learning Using Uncertainty to Weight Losses for Scene Geometry and Semantics" by Kendall, Gal and Cipolla. :param losses: A dict containing all losses that should be merged. :param logging_namespace: Variable scope in which multiloss lives. :param exclude_from_weighting: A list of losses that are already weighted and should not be sigma weighted. :return: A single loss.
entailment
def focus_loss(labels, probs, loss, gamma): """ Calculate the alpha balanced focal loss. See the focal loss paper: "Focal Loss for Dense Object Detection" [by Facebook AI Research] :param labels: A float tensor of shape [batch_size, ..., num_classes] representing the label class probabilities. :param probs: A float tensor of shape [batch_size, ..., num_classes] representing the probs (after softmax). :param loss: A float tensor of shape [batch_size, ...] representing the loss that should be focused. :param gamma: The focus parameter. :return: A tensor representing the weighted cross entropy. """ with tf.variable_scope("focus_loss"): # Compute p_t that is used in paper. # FIXME is it possible that the 1-p term does not make any sense? p_t = tf.reduce_sum(probs * labels, axis=-1)# + tf.reduce_sum((1.0 - probs) * (1.0 - labels), axis=-1) focal_factor = tf.pow(1.0 - p_t, gamma) if gamma > 0 else 1 # Improve stability for gamma = 0 return tf.stop_gradient(focal_factor) * loss
Calculate the alpha balanced focal loss. See the focal loss paper: "Focal Loss for Dense Object Detection" [by Facebook AI Research] :param labels: A float tensor of shape [batch_size, ..., num_classes] representing the label class probabilities. :param probs: A float tensor of shape [batch_size, ..., num_classes] representing the probs (after softmax). :param loss: A float tensor of shape [batch_size, ...] representing the loss that should be focused. :param gamma: The focus parameter. :return: A tensor representing the weighted cross entropy.
entailment
def layer(op): '''Decorator for composable network layers.''' def layer_decorated(self, *args, **kwargs): # Automatically set a name if not provided. name = kwargs.setdefault('name', self.get_unique_name(op.__name__)) # Figure out the layer inputs. if len(self.terminals) == 0: raise RuntimeError('No input variables found for layer %s.' % name) elif len(self.terminals) == 1: layer_input = self.terminals[0] else: layer_input = list(self.terminals) # Perform the operation and get the output. layer_output = op(self, layer_input, *args, **kwargs) # Add to layer LUT. self.layers[name] = layer_output # This output is now the input for the next layer. self.feed(layer_output) # Return self for chained calls. return self return layer_decorated
Decorator for composable network layers.
entailment
def multi_output_layer(op): '''Decorator for composable network layers.''' def layer_decorated(self, *args, **kwargs): # Automatically set a name if not provided. name = kwargs.setdefault('name', self.get_unique_name(op.__name__)) output_names = kwargs.setdefault('output_names', self.get_unique_name(op.__name__)) # Figure out the layer inputs. if len(self.terminals) == 0: raise RuntimeError('No input variables found for layer %s.' % name) elif len(self.terminals) == 1: layer_input = self.terminals[0] else: layer_input = list(self.terminals) # Perform the operation and get the output. layer_output = op(self, layer_input, *args, **kwargs) # Add to layer LUT. for i in range(len(output_names)): self.layers[output_names[i]] = layer_output[i] # This output is now the input for the next layer. self.feed(layer_output) # Return self for chained calls. return self return layer_decorated
Decorator for composable network layers.
entailment
def _load(self, data_path, ignore_missing=False): '''Load network weights. data_path: The path to the numpy-serialized network weights session: The current TensorFlow session ignore_missing: If true, serialized weights for missing layers are ignored. ''' if data_path.endswith(".npz"): data_dict = np.load(data_path) keys = sorted(data_dict.keys()) for i, k in enumerate(keys): data = data_dict[k] op_name = "_".join(k.split("_")[:-1]) param_name = "weights" if k.split("_")[-1] == "W" else "biases" if self.verbose: print("Loaded: {} {}".format(op_name, param_name)) if op_name not in self.weights: self.weights[op_name] = {} self.weights[op_name][param_name] = data elif data_path.endswith(".npy"): data_dict = np.load(data_path).item() for op_name in data_dict: with tf.variable_scope(op_name, reuse=True): for param_name, data in data_dict[op_name].iteritems(): if self.verbose: print("Loaded: {} {}".format(op_name, param_name)) if op_name not in self.weights: self.weights[op_name] = {} self.weights[op_name][param_name] = data else: raise RuntimeError("Invalid file type.")
Load network weights. data_path: The path to the numpy-serialized network weights session: The current TensorFlow session ignore_missing: If true, serialized weights for missing layers are ignored.
entailment
def feed(self, *args): '''Set the input(s) for the next operation by replacing the terminal nodes. The arguments can be either layer names or the actual layers. ''' assert len(args) != 0 self.terminals = [] for fed_layer in args: if isinstance(fed_layer, str): try: fed_layer = self.layers[fed_layer] except KeyError: raise KeyError('Unknown layer name fed: %s' % fed_layer) self.terminals.append(fed_layer) return self
Set the input(s) for the next operation by replacing the terminal nodes. The arguments can be either layer names or the actual layers.
entailment
def get_unique_name(self, prefix): '''Returns an index-suffixed unique name for the given prefix. This is used for auto-generating layer names based on the type-prefix. ''' ident = sum(t.startswith(prefix) for t, _ in self.layers.items()) + 1 return '%s_%d' % (prefix, ident)
Returns an index-suffixed unique name for the given prefix. This is used for auto-generating layer names based on the type-prefix.
entailment
def make_var(self, op_name, name, shape): '''Creates a new TensorFlow variable.''' if op_name in self.weights and name in self.weights[op_name]: if self.verbose: print("Using: {} {}".format(op_name, name)) initializer = tf.constant(self.weights[op_name][name], shape=shape) return tf.get_variable(name, initializer=initializer, trainable=self.trainable) return tf.get_variable(name, shape, trainable=self.trainable)
Creates a new TensorFlow variable.
entailment
def mode_to_str(mode): """ Converts a tf.estimator.ModeKeys in a nice readable string. :param mode: The mdoe as a tf.estimator.ModeKeys :return: A human readable string representing the mode. """ if mode == tf.estimator.ModeKeys.TRAIN: return "train" if mode == tf.estimator.ModeKeys.EVAL: return "eval" if mode == tf.estimator.ModeKeys.PREDICT: return "predict" return "unknown"
Converts a tf.estimator.ModeKeys in a nice readable string. :param mode: The mdoe as a tf.estimator.ModeKeys :return: A human readable string representing the mode.
entailment
def tf_if(condition, a, b): """ Implements an if condition in tensorflow. :param condition: A boolean condition. :param a: Case a. :param b: Case b. :return: A if condition was true, b otherwise. """ int_condition = tf.to_float(tf.to_int64(condition)) return a * int_condition + (1 - int_condition) * b
Implements an if condition in tensorflow. :param condition: A boolean condition. :param a: Case a. :param b: Case b. :return: A if condition was true, b otherwise.
entailment
def _read_data_legacy(prefix, batch_size): """ Loads a tf record as tensors you can use. :param prefix: The path prefix as defined in the write data method. :param batch_size: The batch size you want for the tensors. :return: A feature tensor dict and a label tensor dict. """ prefix = prefix.replace("\\", "/") folder = "/".join(prefix.split("/")[:-1]) phase = prefix.split("/")[-1] config = json.load(open(prefix + '_config.json')) num_threads = config["num_threads"] filenames = [folder + "/" + f for f in listdir(folder) if isfile(join(folder, f)) and phase in f and not "config.json" in f] # Create a tf object for the filename list and the readers. filename_queue = tf.train.string_input_producer(filenames) readers = [_read_tf_record(filename_queue, config) for _ in range(num_threads)] batch_dict = tf.train.shuffle_batch_join( readers, batch_size=batch_size, capacity=10 * batch_size, min_after_dequeue=5 * batch_size ) # Add batch dimension to feature and label shape feature_batch = {} label_batch = {} for k in batch_dict.keys(): shape = tuple([batch_size] + list(config[k]["shape"])) tensor = tf.reshape(batch_dict[k], shape, name="input/"+phase+"/" + k + "_reshape") if "feature_" in k: feature_batch["_".join(k.split("_")[1:])] = tensor if "label_" in k: label_batch["_".join(k.split("_")[1:])] = tensor return feature_batch, label_batch
Loads a tf record as tensors you can use. :param prefix: The path prefix as defined in the write data method. :param batch_size: The batch size you want for the tensors. :return: A feature tensor dict and a label tensor dict.
entailment
def _read_data(prefix, batch_size, augmentation=None): """ Loads a dataset. :param prefix: The path prefix as defined in the write data method. :param batch_size: The batch size you want for the tensors. :param augmentation: An augmentation function. :return: A tensorflow.data.dataset object. """ prefix = prefix.replace("\\", "/") folder = "/".join(prefix.split("/")[:-1]) phase = prefix.split("/")[-1] config = json.load(open(prefix + '_config.json')) num_threads = config["num_threads"] filenames = [folder + "/" + f for f in listdir(folder) if isfile(join(folder, f)) and phase in f and not "config.json" in f] dataset = tf.data.TFRecordDataset(filenames=filenames, num_parallel_reads=num_threads) dataset = dataset.shuffle(buffer_size=10 * batch_size) dataset = dataset.repeat() dataset = dataset.map(map_func=_create_parser_fn(config, phase), num_parallel_calls=num_threads) if augmentation is not None: dataset = dataset.map(map_func=augmentation, num_parallel_calls=num_threads) dataset = dataset.batch(batch_size=batch_size) dataset = dataset.prefetch(buffer_size=1) return dataset
Loads a dataset. :param prefix: The path prefix as defined in the write data method. :param batch_size: The batch size you want for the tensors. :param augmentation: An augmentation function. :return: A tensorflow.data.dataset object.
entailment
def create_input_fn(prefix, batch_size, augmentation=None): """ Loads a dataset. :param prefix: The path prefix as defined in the write data method. :param batch_size: The batch size you want for the tensors. :param augmentation: An augmentation function. :return: An input function for a tf estimator. """ # Check if the version is too old for dataset api to work better than manually loading data. if tf.__version__.startswith("1.6") or tf.__version__.startswith("1.5") or tf.__version__.startswith("1.4") \ or tf.__version__.startswith("1.3") or tf.__version__.startswith("1.2") \ or tf.__version__.startswith("1.1") or tf.__version__.startswith("1.0"): def input_fn(): with tf.variable_scope("input_pipeline"): return _read_data_legacy(prefix, batch_size) return input_fn else: def input_fn(): with tf.variable_scope("input_pipeline"): return _read_data(prefix, batch_size, augmentation) return input_fn
Loads a dataset. :param prefix: The path prefix as defined in the write data method. :param batch_size: The batch size you want for the tensors. :param augmentation: An augmentation function. :return: An input function for a tf estimator.
entailment
def write_data(hyper_params, mode, sequence, num_threads): """ Write a tf record containing a feature dict and a label dict. :param hyper_params: The hyper parameters required for writing {"problem": {"augmentation": {"steps": Int}}} :param mode: The mode specifies the purpose of the data. Typically it is either "train" or "validation". :param sequence: A tf.keras.utils.sequence. :param num_threads: The number of threads. (Recommended: 4 for training and 2 for validation seems to works nice) :return: """ if not isinstance(sequence, Sequence) and not (callable(getattr(sequence, "__getitem__", None)) and callable(getattr(sequence, "__len__", None))): raise ValueError("sequence must be tf.keras.utils.Sequence or a subtype or implement __len__(self) and __getitem__(self, idx)") prefix = os.path.join(hyper_params.train.get("tf_records_path", "tfrecords"), mode) prefix = prefix.replace("\\", "/") data_tmp_folder = "/".join(prefix.split("/")[:-1]) if not os.path.exists(data_tmp_folder): os.makedirs(data_tmp_folder) args = [(hyper_params, sequence, num_threads, i, (prefix + "_%d.tfrecords") % i) for i in range(num_threads)] # Retrieve a single batch sample_feature, sample_label = sequence[0] config = {"num_threads": num_threads} for k in sample_feature.keys(): config["feature_" + k] = {"shape": sample_feature[k].shape[1:], "dtype": sample_feature[k].dtype.name} for k in sample_label.keys(): config["label_" + k] = {"shape": sample_label[k].shape[1:], "dtype": sample_label[k].dtype.name} with open(prefix + '_config.json', 'w') as outfile: json.dump(config, outfile) pool = Pool(processes=num_threads) pool.map(_write_tf_record_pool_helper, args)
Write a tf record containing a feature dict and a label dict. :param hyper_params: The hyper parameters required for writing {"problem": {"augmentation": {"steps": Int}}} :param mode: The mode specifies the purpose of the data. Typically it is either "train" or "validation". :param sequence: A tf.keras.utils.sequence. :param num_threads: The number of threads. (Recommended: 4 for training and 2 for validation seems to works nice) :return:
entailment
def env(self, value): """ Sets context.env, context.env_short, and context.account_alias if env is valid For envs of the form "global.<account>" and "mgmt.<account_alias>", env is captured as "global" or "mgmt" and account_alias is parsed out of the full env rather than looked up Args: value: the fully-qualified env value Raises: ValueError if env is not valid """ env_valid(value) self._env_full = value if value.find(".") == -1: # plain environment, e.g. prod, staging, proto<n> self._env = value self._account_alias = get_account_alias(value) else: # "<env>.<account_alias>" form, e.g. global.ellationeng or mgmt.ellationeng self._env, self._account_alias = value.split(".") # since we extracted an env, must reconfirm that it's legit global_env_valid(self._env) self._env_short = get_env_short(value)
Sets context.env, context.env_short, and context.account_alias if env is valid For envs of the form "global.<account>" and "mgmt.<account_alias>", env is captured as "global" or "mgmt" and account_alias is parsed out of the full env rather than looked up Args: value: the fully-qualified env value Raises: ValueError if env is not valid
entailment
def service_registry(self, sr): """ Sets service registry object in context, doesn't check it Args: sr: EFServiceRegistry object """ if type(sr) is not EFServiceRegistry: raise TypeError("sr value must be type 'EFServiceRegistry'") self._service_registry = sr
Sets service registry object in context, doesn't check it Args: sr: EFServiceRegistry object
entailment
def account_id(self, value): """ Sets the current account id Args: value: current account id (string) Returns: None """ if type(value) is not str: raise TypeError("commit value must be string") self._account_id = value
Sets the current account id Args: value: current account id (string) Returns: None
entailment
def aws_client(self, client_id=None): """ Get AWS client if it exists (must have been formerly stored with set_aws_clients) If client_id is not provided, returns the dictionary of all clients Args: client_id: label for the client, e.g. 'ec2'; omit to get a dictionary of all clients Returns: aws client if found, or None if not """ if client_id is None: return self._aws_clients elif self._aws_clients is not None and self._aws_clients.has_key(client_id): return self._aws_clients[client_id] else: return None
Get AWS client if it exists (must have been formerly stored with set_aws_clients) If client_id is not provided, returns the dictionary of all clients Args: client_id: label for the client, e.g. 'ec2'; omit to get a dictionary of all clients Returns: aws client if found, or None if not
entailment
def set_aws_clients(self, clients): """ Stash a dictionary of AWS clients in the context object Args: clients: dictionary of clients """ if type(clients) is not dict: raise TypeError("clients must be a dict") self._aws_clients = clients
Stash a dictionary of AWS clients in the context object Args: clients: dictionary of clients
entailment
def handle_args_and_set_context(args): """ Args: args: the command line args, probably passed from main() as sys.argv[1:] Returns: a populated EFVersionContext object """ parser = argparse.ArgumentParser() parser.add_argument("service_name", help="name of the service") parser.add_argument("key", help="version key to look up for <service_name> such as 'ami-id' (list in EF_Config)") parser.add_argument("env", help=", ".join(EFConfig.ENV_LIST)) group = parser.add_mutually_exclusive_group(required=True) group.add_argument("--get", help="get current version", action="store_true") group.add_argument("--set", help="set current version of <key> to <value> for <service_name>") group.add_argument("--rollback", help="set current version to most recent 'stable' version in history", action="store_true") group.add_argument("--rollback-to", help="rollback current version to <ami-id> in history", action="store", metavar='<ami-id>') group.add_argument("--history", help="Show version history for env/service/key", choices=['json', 'text']) group.add_argument("--show", help="Show keys and values. '*' allowed for <key> and <env>", action="store_true", default=False) parser.add_argument("--build", help="On --set, also set the externally defined build number associated with the version entity", default="") parser.add_argument("--commit_hash", help="On --set, also set the commit hash associated with the version entity", default="") parser.add_argument("--commit", help="Actually --set or --rollback (dry run if omitted)", action="store_true", default=False) parser.add_argument("--devel", help="Allow running from branch; don't refresh from origin", action="store_true", default=False) parser.add_argument("--force_env_full", help="Override env with env_full for account-scoped environments", action="store_true", default=False) parser.add_argument("--limit", help="Limit 'history', 'rollback', 'show' to first N records (default 100, max 1000)", type=int, default=100) parser.add_argument("--location", help="On --set, also mark the url location of the static build's version file to" "support dist-hash precheck", default="") if EFConfig.ALLOW_EF_VERSION_SKIP_PRECHECK: parser.add_argument("--noprecheck", help="--set or --rollback without precheck", action="store_true", default=False) parser.add_argument("--sr", help="optional /path/to/service_registry_file.json", default=None) parser.add_argument("--stable", help="On --set, also mark the version 'stable'", action="store_true") parser.add_argument("--verbose", help="Print additional info", action="store_true", default=False) # parse parsed_args = vars(parser.parse_args(args)) context = EFVersionContext() # marshall the inherited context values context._build_number = parsed_args["build"] context._commit_hash = parsed_args["commit_hash"] context.commit = parsed_args["commit"] context.devel = parsed_args["devel"] context._force_env_full = parsed_args["force_env_full"] try: context.env = parsed_args["env"] except ValueError as e: fail("Error in env: {}".format(e.message)) # marshall this module's additional context values context._get = parsed_args["get"] context._history = parsed_args["history"] context._key = parsed_args["key"] if EFConfig.ALLOW_EF_VERSION_SKIP_PRECHECK: context._noprecheck = parsed_args["noprecheck"] if not 1 <= parsed_args["limit"] <= 1000: fail("Error in --limit. Valid range: 1..1000") context._limit = parsed_args["limit"] context._location = parsed_args["location"] context._rollback = parsed_args["rollback"] context._rollback_to = parsed_args["rollback_to"] context._service_name = parsed_args["service_name"] context._show = parsed_args["show"] context._stable = parsed_args["stable"] context._value = parsed_args["set"] # Set up service registry and policy template path which depends on it context.service_registry = EFServiceRegistry(parsed_args["sr"]) # VERBOSE is global global VERBOSE VERBOSE = parsed_args["verbose"] validate_context(context) return context
Args: args: the command line args, probably passed from main() as sys.argv[1:] Returns: a populated EFVersionContext object
entailment
def validate_context(context): """ Set the key for the current context. Args: context: a populated EFVersionContext object """ # Service must exist in service registry if not context.service_registry.service_record(context.service_name): fail("service: {} not found in service registry: {}".format( context.service_name, context.service_registry.filespec)) service_type = context.service_registry.service_record(context.service_name)["type"] # Key must be valid if context.key not in EFConfig.VERSION_KEYS: fail("invalid key: {}; see VERSION_KEYS in ef_config for supported keys".format(context.key)) # Lookup allowed key for service type if "allowed_types" in EFConfig.VERSION_KEYS[context.key] and \ service_type not in EFConfig.VERSION_KEYS[context.key]["allowed_types"]: fail("service_type: {} is not allowed for key {}; see VERSION_KEYS[KEY]['allowed_types']" "in ef_config and validate service registry entry".format(service_type, context.key)) return True
Set the key for the current context. Args: context: a populated EFVersionContext object
entailment