import numpy as np import tensorflow as tf import tensorflow_datasets as tfds import matplotlib.pyplot as plt import time from tensorflow.keras.preprocessing.text import Tokenizer from tensorflow.keras.preprocessing.sequence import pad_sequences from tensorflow.keras import backend as K imdb, info = tfds.load("imdb_reviews", with_info=True, as_supervised=True) vocab_size = 10000 oov_tok = "" padding_type = "post" trunc_type = "post" max_length = 120 num_epochs = 50 embedding_dim = 16 train_data, test_data = imdb['train'], imdb['test'] training_sentences = [] training_labels = [] testing_sentences = [] testing_labels = [] for s,l in train_data: training_sentences.append(str(s.numpy())) training_labels.append(l.numpy()) for s,l in test_data: testing_sentences.append(str(s.numpy())) testing_labels.append(l.numpy()) training_labels_final = np.array(training_labels) testing_labels_final = np.array(testing_labels) tokenizer = Tokenizer(num_words=vocab_size, oov_token=oov_tok) tokenizer.fit_on_texts(training_sentences) word_index = tokenizer.word_index training_sequences = tokenizer.texts_to_sequences(training_sentences) training_padded = pad_sequences(training_sequences, maxlen=max_length, padding=padding_type, truncating=trunc_type) testing_sequences = tokenizer.texts_to_sequences(testing_sentences) testing_padded = pad_sequences(testing_sequences, maxlen=max_length, padding=padding_type, truncating=trunc_type) def create_model(name): if name == 'simple': return tf.keras.Sequential([ tf.keras.layers.Embedding(vocab_size, embedding_dim, input_length=max_length), tf.keras.layers.Flatten(), tf.keras.layers.Dense(6, activation='relu'), tf.keras.layers.Dense(1, activation='sigmoid') ]) elif name == 'lstm': return tf.keras.Sequential([ tf.keras.layers.Embedding(vocab_size, embedding_dim, input_length=max_length), tf.keras.layers.Bidirectional(tf.keras.layers.LSTM(32)), tf.keras.layers.Dense(6, activation='relu'), tf.keras.layers.Dense(1, activation='sigmoid') ]) elif name == 'gru': return tf.keras.Sequential([ tf.keras.layers.Embedding(vocab_size, embedding_dim, input_length=max_length), tf.keras.layers.Bidirectional(tf.keras.layers.GRU(32)), tf.keras.layers.Dense(6, activation='relu'), tf.keras.layers.Dense(1, activation='sigmoid') ]) elif name == 'conv': return tf.keras.Sequential([ tf.keras.layers.Embedding(vocab_size, embedding_dim, input_length=max_length), tf.keras.layers.Conv1D(128, 5, activation='relu'), tf.keras.layers.GlobalAveragePooling1D(), tf.keras.layers.Dense(6, activation='relu'), tf.keras.layers.Dense(1, activation='sigmoid') ]) model_names = ['simple', 'conv', 'lstm', 'gru'] model_stats = {} for model_name in model_names: model = create_model(model_name) model.summary() trainable_param_count = int( np.sum([K.count_params(p) for p in model.trainable_weights])) model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy']) t0 = time.time() history = model.fit( training_padded, training_labels_final, epochs=num_epochs, validation_data=(testing_padded, testing_labels_final)) dt = time.time() - t0 model_stats[model_name] = { 'trainable_param_count': trainable_param_count, 'history': history.history, 'training_time': dt } tf.keras.backend.clear_session() np.save(f"{model_name}.npy", model_stats[model_name]) # Print results fig = plt.figure(figsize=(10.0, 10.0)) plt.title("Comparison of different network archituectures on imdb dataset") plt.subplot(3, 2, 1) plt.title('Training Accuracy') for model_name, stats in model_stats.items(): print(model_name) print(stats['history']) plt.plot(stats['history']['accuracy'], label=model_name) plt.xlabel("Epochs") plt.ylabel("Accuracy") plt.ylim([0.75, 1.0]) plt.legend() plt.subplot(3, 2, 2) plt.title('Validation Accuracy') for model_name, stats in model_stats.items(): print(model_name) print(stats['history']) plt.plot(stats['history']['val_accuracy'], label=model_name) plt.xlabel("Epochs") plt.ylabel("Accuracy") plt.legend() plt.ylim([0.75, 1.0]) plt.subplot(3, 2, 3) plt.title('Training Loss') for model_name, stats in model_stats.items(): print(model_name) print(stats['history']) plt.plot(stats['history']['loss'], label=model_name) plt.xlabel("Epochs") plt.ylabel("Loss") plt.ylim([ 0.0, 3.5]) plt.legend() plt.subplot(3, 2, 4) plt.title('Validation Loss') for model_name, stats in model_stats.items(): print(model_name) print(stats['history']) plt.plot(stats['history']['val_loss'], label=model_name) plt.xlabel("Epochs") plt.ylabel("Loss") plt.legend() plt.ylim([ 0.0, 3.5]) plt.subplot(3, 2, 5) data = [s['training_time'] / 50 for n, s in model_stats.items()] names = [n for n, s in model_stats.items()] plt.title('Training time') X = np.arange(len(names)) rects = plt.bar(X, data, align='center') for val, rect in zip(data, rects): height = rect.get_height() plt.gca().text(rect.get_x() + rect.get_width()/2., height+0.0125, '{:.2f}'.format(val), ha='center', va='bottom') plt.xticks(X, names, rotation=30., ha='right') ylim_upper = np.ceil(max(data) * 1000.0) / 1000.0 + 10.0 plt.ylim([0, ylim_upper]) plt.ylabel('Training time [s] / Epoch') plt.subplot(3, 2, 6) data = [s['trainable_param_count'] for n, s in model_stats.items()] names = [n for n, s in model_stats.items()] plt.title('Parameter count') X = np.arange(len(names)) rects = plt.bar(X, data, align='center') for val, rect in zip(data, rects): height = rect.get_height() plt.gca().text(rect.get_x() + rect.get_width()/2., height+0.0125, '{:.2f}'.format(val), ha='center', va='bottom') plt.xticks(X, names, rotation=30., ha='right') ylim_upper = np.ceil(max(data) * 1000000.0) / 1000000.0 + 100000.0 plt.ylim([0, ylim_upper]) plt.ylabel('#Parameter') plt.tight_layout() plt.savefig("plot.svg") plt.show()