Source Code/notebooks/modeling.py

# -*- coding: utf-8 -*-
"""modeling.ipynb

Automatically generated by Colaboratory.

Original file is located at
    https://colab.research.google.com/drive/1x78fRDZAuK5FaSTKHPGy8eSbZ_gYAFr6
"""

from google.colab import drive
drive.mount('/content/drive')

#!pip install -qqq h5py

#!pip install --upgrade -qqq gensim

!python -m spacy download en_core_web_lg

!pip install -U SpaCy==2.2.0

## Import required libraries

## warnings
import warnings
warnings.filterwarnings("ignore")

## for data
import numpy as np
import pandas as pd

## for plotting
import matplotlib.pyplot as plt
import seaborn as sns

## TF-IDF 
from sklearn.feature_extraction.text import TfidfVectorizer

## T-Sne
from yellowbrick.text import TSNEVisualizer
from sklearn import manifold

## Train-Test Split
from sklearn.model_selection import train_test_split

## Feature selection
from sklearn import feature_selection

## libraraies for classification
from sklearn.pipeline import Pipeline
import sklearn.metrics as skm
from sklearn.metrics import confusion_matrix, accuracy_score
from sklearn.linear_model import LogisticRegression
from sklearn.neighbors import KNeighborsClassifier
from sklearn.svm import SVC
from sklearn.tree import DecisionTreeClassifier
from sklearn.neural_network import MLPClassifier
from sklearn.ensemble import RandomForestClassifier

## for saving model
import pickle

## for explainer
#from lime import lime_text

## detokenization
from nltk.tokenize.treebank import TreebankWordDetokenizer

## for word embedding with gensim
import gensim
import gensim.downloader as gensim_api
from gensim.models import Word2Vec
from gensim.models import KeyedVectors
from keras.preprocessing.text import Tokenizer
from keras.preprocessing.sequence import pad_sequences

## for word embedding with Spacy
import spacy
import en_core_web_lg

## for deep learning
from keras.models import load_model
from keras.models import Model, Sequential
from keras.callbacks import EarlyStopping, ModelCheckpoint
from keras.layers import Conv1D, Dense, Input, LSTM, Embedding, Dropout, Activation, MaxPooling1D
from tensorflow.keras import models, layers, preprocessing as kprocessing
from tensorflow.keras import backend as K
import tensorflow as tf
import keras
from keras.layers import Lambda
import tensorflow as tf
from keras.models import model_from_json

## for bert language model
#import transformers

"""## Loading the dataset:"""

df_all = pd.read_csv("/content/drive/MyDrive/NLP/Depression_Detection/data_cleaning/processed_data/processed_data.csv",
                     sep='\t', encoding='utf-8')

df_all

"""## Classification models as well as LSTM with pretrained model(Spacy):

In order to run a supervised learning model, we first need to convert the clean_text into feature representation.
"""

nlp = en_core_web_lg.load()

## word-embedding
all_vectors = pd.np.array([pd.np.array([token.vector for token in nlp(s)]).mean(axis=0) * pd.np.ones((300)) \
                           for s in df_all['clean_text']])

# split out validation dataset for the end
Y= df_all["label"]
X = all_vectors

from sklearn.model_selection import train_test_split, KFold, cross_val_score, GridSearchCV
validation_size = 0.3
seed = 7
X_train, X_test, Y_train, Y_test = train_test_split(X, Y, test_size=validation_size, random_state=seed)

# test options for classification
num_folds = 10
seed = 7
scoring = 'accuracy'

## spot check the algorithms
models = []
models.append(('LR', LogisticRegression()))
models.append(('KNN', KNeighborsClassifier()))
models.append(('CART', DecisionTreeClassifier()))
models.append(('SVM', SVC()))
## Neural Network
models.append(('NN', MLPClassifier()))
## Ensable Models 
models.append(('RF', RandomForestClassifier()))

## Running the classification models
results = []
names = []
kfold_results = []
test_results = []
train_results = []
for name, model in models:
    kfold = KFold(n_splits=num_folds, random_state=seed)
    cv_results = cross_val_score(model, X_train, Y_train, cv=kfold, scoring=scoring)
    results.append(cv_results)
    names.append(name)
    #msg = "%s: %f (%f)" % (name, cv_results.mean(), cv_results.std())
    #print(msg)
   
   # Full Training period
    res = model.fit(X_train, Y_train)
    train_result = accuracy_score(res.predict(X_train), Y_train)
    train_results.append(train_result)
    
    # Test results
    test_result = accuracy_score(res.predict(X_test), Y_test)
    test_results.append(test_result)    
    
    msg = "%s: %f (%f) %f %f" % (name, cv_results.mean(), cv_results.std(), train_result, test_result)
    print(msg)
    print(confusion_matrix(res.predict(X_test), Y_test))
    #print(classification_report(res.predict(X_test), Y_test))

# compare algorithms
from matplotlib import pyplot
fig = pyplot.figure()
ind = np.arange(len(names))  # the x locations for the groups
width = 0.35  # the width of the bars
fig.suptitle('Algorithm Comparison')
ax = fig.add_subplot(111)
pyplot.bar(ind - width/2, train_results,  width=width, label='Train Error')
pyplot.bar(ind + width/2, test_results, width=width, label='Test Error')
fig.set_size_inches(15,8)
pyplot.legend()
ax.set_xticks(ind)
ax.set_xticklabels(names)
pyplot.show()

"""The best model with the highest accuracy is **Support Vector Machine(SVM)** with **85.79**% accuracy on test dataset. Logistic Regression performed good as well but we see overfitting problem with CART, NN and RF.

### LSTM model:
"""

### Create sequence
vocabulary_size = 20000
tokenizer = Tokenizer(num_words= vocabulary_size)
tokenizer.fit_on_texts(df_all['clean_text'])
sequences = tokenizer.texts_to_sequences(df_all['clean_text'])
X_LSTM = pad_sequences(sequences, maxlen=50)

## Split the data into train and test
Y_LSTM = df_all["label"]
X_train_LSTM, X_test_LSTM, Y_train_LSTM, Y_test_LSTM = train_test_split(X_LSTM, \
                       Y_LSTM, test_size=validation_size, random_state=seed)

from keras.wrappers.scikit_learn import KerasClassifier
def create_model(input_length=50):
    model = Sequential()
    model.add(Embedding(20000, 300, input_length=50))
    model.add(LSTM(100, dropout=0.2, recurrent_dropout=0.2))
    model.add(Dense(1, activation='sigmoid'))
    model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])    
    return model    
model_LSTM = KerasClassifier(build_fn=create_model, epochs=3, verbose=1, validation_split=0.4)
model_LSTM.fit(X_train_LSTM, Y_train_LSTM)

train_result_LSTM = accuracy_score(model_LSTM.predict(X_train_LSTM), Y_train_LSTM)
# Test results
test_result_LSTM = accuracy_score(model_LSTM.predict(X_test_LSTM), Y_test_LSTM)

print("train result:", train_result_LSTM)
print("test result:", test_result_LSTM)

confusion_matrix(model_LSTM.predict(X_test_LSTM), Y_test_LSTM)

"""### Compare all the models:"""

train_results.append(train_result_LSTM);test_results.append(test_result_LSTM)
names.append("LSTM")

# compare algorithms
from matplotlib import pyplot
fig = pyplot.figure()
ind = np.arange(len(names))  # the x locations for the groups
width = 0.35  # the width of the bars
fig.suptitle('Algorithm Comparison')
ax = fig.add_subplot(111)
pyplot.bar(ind - width/2, train_results,  width=width, label='Train Error')
pyplot.bar(ind + width/2, test_results, width=width, label='Test Error')
fig.set_size_inches(15,8)
pyplot.legend()
ax.set_xticks(ind)
ax.set_xticklabels(names)
pyplot.show()
plt.savefig('/content/drive/MyDrive/NLP/Depression_Detection/modeling/classification_comparision.png')

"""## Evaluate the performance:

*   **Accuracy:** the fraction of predictions the model got right.
*   **Confusion Matrix:** a summary table that breaks down the number of correct and incorrect predictions by each class.
*   **ROC:** a plot that illustrates the true positive rate against the false positive rate at various threshold settings. The area under the curve (AUC) indicates the probability that the classifier will rank a randomly chosen positive observation higher than a randomly chosen negative one.
*  **Precision:** the fraction of relevant instances among the retrieved instances.
*  **Recall:** the fraction of the total amount of relevant instances that were actually retrieved.
"""

def conf_matrix_acc(y_true, y_pred):
  ## Plot confusion matrix
  cm = confusion_matrix(y_true, y_pred)
  fig, ax = plt.subplots()
  sns.heatmap(cm, annot=True, fmt='d', ax=ax, cmap=plt.cm.Blues, 
            cbar=False)
  ax.set(xlabel="Pred", ylabel="True", xticklabels=classes, 
       yticklabels=classes, title="Confusion matrix")
  plt.yticks(rotation=0)
  print("=========================================")
  print(f'Accuracy score is : {accuracy_score(y_true, y_pred)}')
  print("=========================================")
  print("Detail:")
  print(skm.classification_report(y_true, y_pred))

## Plot ROC and precision-recall curve
def roc_precision_auc():
  fig, ax = plt.subplots(nrows=1, ncols=2)
  ## Plot roc
  for i in range(len(classes)):
      fpr, tpr, thresholds = skm.roc_curve(y_test_array[:,i],  
                            probs[:,i])
      ax[0].plot(fpr, tpr, lw=3, 
                label='{0} (area={1:0.2f})'.format(classes[i], 
                                skm.auc(fpr, tpr))
                 )
  ax[0].plot([0,1], [0,1], color='navy', lw=3, linestyle='--')
  ax[0].set(xlim=[-0.05,1.0], ylim=[0.0,1.05], 
            xlabel='False Positive Rate', 
            ylabel="True Positive Rate (Recall)", 
            title="Receiver operating characteristic")
  ax[0].legend(loc="lower right")
  ax[0].grid(True)

  ## Plot precision-recall curve
  for i in range(len(classes)):
    precision, recall, thresholds = skm.precision_recall_curve(
                y_test_array[:,i], probs[:,i])
    ax[1].plot(recall, precision, lw=3, 
               label='{0} (area={1:0.2f})'.format(classes[i], 
                                  skm.auc(recall, precision))
              )
  ax[1].set(xlim=[0.0,1.05], ylim=[0.0,1.05], xlabel='Recall', 
            ylabel="Precision", title="Precision-Recall curve")
  ax[1].legend(loc="best")
  ax[1].grid(True)
  plt.show()
  #plt.savefig('/content/drive/MyDrive/NLP/Depression_Detection/modeling/ROC_Precision_LR.png')
  #plt.savefig('/content/drive/MyDrive/NLP/Depression_Detection/modeling/ROC_Precision_SVM.png')  
  ## AUC score
  print(f'AUC score is : {skm.roc_auc_score(Y_test, probs[:,1])}')

"""## Support Vector Machine(SVM) with word embedding:"""

nlp = en_core_web_lg.load()

## word-embedding
all_vectors = pd.np.array([pd.np.array([token.vector for token in nlp(s)]).mean(axis=0) * pd.np.ones((300)) \
                           for s in df_all['clean_text']])

# split out validation dataset for the end
Y= df_all["label"]
X = all_vectors

from sklearn.model_selection import train_test_split, KFold, cross_val_score, GridSearchCV
validation_size = 0.3
seed = 7
X_train, X_test, Y_train, Y_test = train_test_split(X, Y, test_size=validation_size, random_state=seed)

# test options for classification
num_folds = 10
seed = 7
scoring = 'accuracy'

#Create a svm Classifier
clf = SVC(probability=True)

## Running the svm Classifier
   
# Full Training period
res = clf.fit(X_train, Y_train)
train_result = accuracy_score(res.predict(X_train), Y_train)
test_result = accuracy_score(res.predict(X_test), Y_test)

print("train_result:", "test_resuld:", train_result, test_result, sep=" ")

## Save the Modle to file in the current working directory
SVM = "/content/drive/MyDrive/NLP/Depression_Detection/modeling/model_svm1.pkl"  

with open(SVM, 'wb') as file:  
    pickle.dump(clf, file)

## Load the Model back from file
with open(SVM, 'rb') as file:  
    clf = pickle.load(file)

clf

## Test results
## 
y_pred_svm = res.predict(X_test) 
classes = np.unique(Y_test.to_list())
y_test_array = pd.get_dummies(Y_test, drop_first=False).values
probs = res.predict_proba(X_test)
conf_matrix_acc(Y_test.to_list(),y_pred_svm)
roc_precision_auc()

"""## Exploring False positive and False negative:"""

## creating lists of true values and predictions
y_test_1 = [x for x in y_test]
y_pred_lr_1 = [x for x in y_pred_lr]

## Find the indices of wrong predictions
idx = [] 
for i in range(len(y_test_1)):
  if y_test_1[i] != y_pred_lr_1[i]:
    idx.append(i)
    i+=1

print('There are", {} "wrong preditions", len(idx))

wrong_arr = cv.inverse_transform(X_test_tfidf[idx])

## detokenize the wrong array
detokenized = [TreebankWordDetokenizer().detokenize(x) for x in wrong_arr]

detokenized[:50]

"""There is no specific patterns between false positive and false negative predictions."""