Datasets:
jblitzar
/
github-python

Dataset card Data Studio Files
xet
Community
text
stringlengths
0
93.6k
same_class_idx = np.where((data_xy[1] == data_class))[0]
diff_class_idx = np.where(data_xy[1] != data_class)[0]
A_P_pairs = random.sample(list(permutations(same_class_idx, 2)), k=ap_pairs) # Generating Anchor-Positive pairs
Neg_idx = random.sample(list(diff_class_idx), k=an_pairs)
# train
A_P_len = len(A_P_pairs)
#Neg_len = len(Neg_idx)
for ap in A_P_pairs[:int(A_P_len * trainsize)]:
Anchor = data_xy[0][ap[0]]
y_Anchor = data_xy[1][ap[0]]
Positive = data_xy[0][ap[1]]
y_Pos = data_xy[1][ap[1]]
for n in Neg_idx:
Negative = data_xy[0][n]
y_Neg = data_xy[1][n]
triplet_train_pairs.append([Anchor, Positive, Negative])
y_triplet_pairs.append([y_Anchor, y_Pos, y_Neg])
# test
return np.array(triplet_train_pairs), np.array(y_triplet_pairs)
def triplet_loss(y_true, y_pred, alpha=0.4):
"""
Implementation of the triplet loss function
Arguments:
y_true -- true labels, required when you define a loss in Keras, you don't need it in this function.
y_pred -- python list containing three objects:
anchor -- the encodings for the anchor data
positive -- the encodings for the positive data (similar to anchor)
negative -- the encodings for the negative data (different from anchor)
Returns:
loss -- real number, value of the loss
"""
print('y_pred.shape = ', y_pred)
total_lenght = y_pred.shape.as_list()[-1]
# print('total_lenght=', total_lenght)
# total_lenght =12
anchor = y_pred[:, 0:int(total_lenght * 1 / 3)]
positive = y_pred[:, int(total_lenght * 1 / 3):int(total_lenght * 2 / 3)]
negative = y_pred[:, int(total_lenght * 2 / 3):int(total_lenght * 3 / 3)]
# distance between the anchor and the positive
pos_dist = K.sum(K.square(anchor - positive), axis=1)
# distance between the anchor and the negative
neg_dist = K.sum(K.square(anchor - negative), axis=1)
# compute loss
basic_loss = pos_dist - neg_dist + alpha
loss = K.maximum(basic_loss, 0.0)
return loss
def triplet_center_loss(y_true, y_pred, n_classes= 10, alpha=0.4):
"""
Implementation of the triplet loss function
Arguments:
y_true -- true labels, required when you define a loss in Keras, you don't need it in this function.
y_pred -- python list containing three objects:
anchor -- the encodings for the anchor data
positive -- the encodings for the positive data (similar to anchor)
negative -- the encodings for the negative data (different from anchor)
Returns:
loss -- real number, value of the loss
"""
print('y_pred.shape = ', y_pred)
total_lenght = y_pred.shape.as_list()[-1]
# print('total_lenght=', total_lenght)
# total_lenght =12
# repeat y_true for n_classes and == np.arange(n_classes)
# repeat also y_pred and apply mask
# obtain min for each column min vector for each class
classes = tf.range(0, n_classes,dtype=tf.float32)
y_pred_r = tf.reshape(y_pred, (tf.shape(y_pred)[0], 1))
y_pred_r = tf.keras.backend.repeat(y_pred_r, n_classes)
y_true_r = tf.reshape(y_true, (tf.shape(y_true)[0], 1))
y_true_r = tf.keras.backend.repeat(y_true_r, n_classes)
mask = tf.equal(y_true_r[:, :, 0], classes)
#mask2 = tf.ones((tf.shape(y_true_r)[0], tf.shape(y_true_r)[1])) # todo inf
# use tf.where(tf.equal(masked, 0.0), np.inf*tf.ones_like(masked), masked)
masked = y_pred_r[:, :, 0] * tf.cast(mask, tf.float32) #+ (mask2 * tf.cast(tf.logical_not(mask), tf.float32))*tf.constant(float(2**10))
masked = tf.where(tf.equal(masked, 0.0), np.inf*tf.ones_like(masked), masked)
minimums = tf.math.reduce_min(masked, axis=1)
loss = K.max(y_pred - minimums +alpha ,0)