ium_464979/sacred/sacred_training_model.py

import pandas as pd
from tensorflow.keras.preprocessing.text import Tokenizer
from tensorflow.keras.preprocessing.sequence import pad_sequences
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Embedding, GlobalAveragePooling1D, Dense
from sklearn.metrics import accuracy_score, precision_recall_fscore_support, mean_squared_error
from sacred import Experiment
from sacred.observers import MongoObserver, FileStorageObserver
from math import sqrt

ex = Experiment('464979')
ex.observers.append(MongoObserver(url='mongodb://admin:IUM_2021@tzietkiewicz.vm.wmi.amu.edu.pl:27017'))
ex.observers.append(FileStorageObserver('sacred_runs'))

@ex.config
def my_config():
    epochs = 10
    batch_size = 32

@ex.automain
def run_experiment(epochs, batch_size, _run):
    train_data = pd.read_csv('beer_reviews_train.csv')
    X_train = train_data[['review_aroma', 'review_appearance', 'review_palate', 'review_taste']]
    y_train = train_data['review_overall']

    tokenizer = Tokenizer(num_words=10000)
    tokenizer.fit_on_texts(X_train)
    X_train_seq = tokenizer.texts_to_sequences(X_train)

    X_train_pad = pad_sequences(X_train_seq, maxlen=100)

    model = Sequential([
        Embedding(input_dim=10000, output_dim=16, input_length=100),
        GlobalAveragePooling1D(),
        Dense(16, activation='relu'),
        Dense(1, activation='sigmoid')
    ])

    model.compile(optimizer='adam',
                  loss='binary_crossentropy',
                  metrics=['accuracy'])

    model.fit(X_train_pad, y_train, epochs=epochs, batch_size=batch_size, validation_split=0.1)

    model.save('beer_review_sentiment_model.keras')
    _run.add_artifact('beer_review_sentiment_model.keras')

    test_data = pd.read_csv('beer_reviews_test.csv')
    X_test = test_data[['review_aroma', 'review_appearance', 'review_palate', 'review_taste']]
    y_test = test_data['review_overall']

    tokenizer = Tokenizer(num_words=10000)
    tokenizer.fit_on_texts(X_test)

    X_test_text = X_test.astype(str).agg(' '.join, axis=1)
    X_test_seq = tokenizer.texts_to_sequences(X_test_text)
    X_test_pad = pad_sequences(X_test_seq, maxlen=100)

    predictions = model.predict(X_test_pad)

    if len(predictions.shape) > 1:
        predictions = predictions[:, 0]

    results = pd.DataFrame({'Predictions': predictions, 'Actual': y_test})
    results.to_csv('beer_review_sentiment_predictions.csv', index=False)

    y_pred = results['Predictions']
    y_test = results['Actual']
    y_test_binary = (y_test >= 3).astype(int)

    accuracy = accuracy_score(y_test_binary, y_pred.round())
    precision, recall, f1, _ = precision_recall_fscore_support(y_test_binary, y_pred.round(), average='micro')
    rmse = sqrt(mean_squared_error(y_test, y_pred))

    print(f'Accuracy: {accuracy}')
    print(f'Micro-avg Precision: {precision}')
    print(f'Micro-avg Recall: {recall}')
    print(f'F1 Score: {f1}')
    print(f'RMSE: {rmse}')

    _run.add_resource('./beer_reviews_train.csv')
    _run.add_resource('./beer_reviews_test.csv')

    return accuracy