ium_444452/Scripts/train_neural_network.py

#!/usr/bin/python
import os
import pprint
import sys
import pandas as pd
from keras.models import Sequential, load_model
from keras import layers
from keras.preprocessing.text import Tokenizer
from keras.preprocessing.sequence import pad_sequences
from sklearn.metrics import precision_score, recall_score, f1_score, accuracy_score
import logging

logging.getLogger("tensorflow").setLevel(logging.ERROR)


def tokenize(x, x_train, x_test, max_len):
    tokenizer = Tokenizer(num_words=14000)
    tokenizer.fit_on_texts(x)
    train_x = tokenizer.texts_to_sequences(x_train)
    test_x = tokenizer.texts_to_sequences(x_test)
    vocabulary_length = len(tokenizer.word_index) + 1

    train_x = pad_sequences(train_x, padding='post', maxlen=max_len)
    test_x = pad_sequences(test_x, padding='post', maxlen=max_len)
    return train_x, test_x, vocabulary_length


def evaluate_and_save(model, x, y, abs_path):
    loss, accuracy = model.evaluate(x, y, verbose=False)
    y_predicted = (model.predict(x) >= 0.5).astype(int)
    with open(os.path.join(abs_path, 'neural_network_evaluation.txt'), "w") as log_file:
        for obj in (
                ('Accuracy: ', accuracy), ('Loss: ', loss), ('Precision: ', precision_score(y, y_predicted)),
                ('Recall: ', recall_score(y, y_predicted)), ('F1: ', f1_score(y, y_predicted)),
                ('Accuracy: ', accuracy_score(y, y_predicted))):
            pprint.pprint(obj, log_file)


def load_trained_model(abs_path, model_name):
    return load_model(os.path.join(abs_path, model_name))


def save_model(model, abs_path, model_name):
    model.save(os.path.join(abs_path, model_name), save_format='h5')


def train_model(model, x_train, y_train):
    model.fit(x_train, y_train, epochs=1, verbose=False, batch_size=50)


def get_model(output_dim, vocabulary_length):
    model = Sequential()
    model.add(layers.Embedding(input_dim=vocabulary_length,
                               output_dim=output_dim,
                               input_length=100))
    model.add(layers.Flatten())
    model.add(layers.Dense(10, activation='relu'))
    model.add(layers.Dense(1, activation='sigmoid'))
    model.compile(optimizer='adam',
                  loss='binary_crossentropy',
                  metrics=['accuracy'])
    return model


def split_data(data):
    x = data['tokens']
    y = data['fraudulent']
    return x, y


def load_data(data_path, filename) -> pd.DataFrame:
    return pd.read_csv(os.path.join(data_path, filename))


def main():
    data_path = sys.argv[1]
    abs_data_path = os.path.abspath(data_path)
    train_data = load_data(abs_data_path, 'train_data.csv')
    test_data = load_data(abs_data_path, 'test_data.csv')
    x_train, y_train = split_data(train_data)
    x_test, y_test = split_data(test_data)
    x_train, x_test, vocab_size = tokenize(pd.concat([x_train, x_test]), x_train, x_test, 100)
    model = get_model(50, vocab_size)
    train_model(model, x_train, y_train)
    save_model(model, abs_data_path, 'neural_network_model')
    evaluate_and_save(model, x_test, y_test, abs_data_path)


if __name__ == '__main__':
    main()