import tensorflow as tf from tensorflow.keras.models import Sequential from tensorflow.keras.layers import Dense, InputLayer, Dropout, Conv1D, Conv2D, Flatten, Reshape, MaxPooling1D, MaxPooling2D, AveragePooling2D, BatchNormalization, Permute, ReLU, Softmax from tensorflow.keras.optimizers.legacy import Adam EPOCHS = 200 LEARNING_RATE = args.learning_rate or 0.0005 # If True, non-deterministic functions (e.g. shuffling batches) are not used. # This is False by default. ENSURE_DETERMINISM = args.ensure_determinism # this controls the batch size, or you can manipulate the tf.data.Dataset objects yourself BATCH_SIZE = args.batch_size or 64 if not ENSURE_DETERMINISM: train_dataset = train_dataset.shuffle(buffer_size=BATCH_SIZE*4) train_dataset=train_dataset.batch(BATCH_SIZE, drop_remainder=False) validation_dataset = validation_dataset.batch(BATCH_SIZE, drop_remainder=False) import tensorflow as tf from tensorflow.keras import layers, models from tensorflow.keras.regularizers import l2 # Assuming you're using a regularization factor of 0.001 for L2 regularization reg = None def simplified_residual_block(x, filters, kernel_size=3, stride=1, dropout_rate=0.0): # Shortcut adjustment if needed shortcut = x if stride != 1 or x.shape[-1] != filters: shortcut = layers.Conv1D(filters, 1, strides=stride, padding='same', use_bias=False, kernel_regularizer=reg)(shortcut) shortcut = layers.BatchNormalization()(shortcut) # First convolution x = layers.Conv1D(filters, kernel_size=kernel_size, strides=stride, padding='same', use_bias=False, kernel_regularizer=reg)(x) x = layers.BatchNormalization()(x) x = layers.Activation('relu')(x) if dropout_rate > 0: x = layers.Dropout(dropout_rate)(x) # Add shortcut and final activation x = layers.add([x, shortcut]) x = layers.Activation('relu')(x) return x def build_small_resnet_18(input_shape=(1800, 1), num_classes=10, dropout_rate=0.0): inputs = tf.keras.Input(shape=input_shape) reshaped_inputs = tf.keras.layers.Reshape((300, 6))(inputs) # Reshape to (300, 6) # Initial Convolution x = layers.Conv1D(16, kernel_size=7, strides=2, padding='same', use_bias=False)(reshaped_inputs) x = layers.BatchNormalization()(x) x = layers.Activation('relu')(x) x = layers.MaxPooling1D(pool_size=3, strides=2, padding='same')(x) # Simplified Residual blocks with fewer filters x = simplified_residual_block(x, 32, dropout_rate=dropout_rate) x = simplified_residual_block(x, 64, stride=2, dropout_rate=dropout_rate) x = simplified_residual_block(x, 128, stride=2, dropout_rate=dropout_rate) # x = simplified_residual_block(x, 256, stride=2, dropout_rate=dropout_rate) # Global Average Pooling and Output x = layers.GlobalAveragePooling1D()(x) outputs = layers.Dense(num_classes, activation='softmax', kernel_regularizer=reg)(x) model = models.Model(inputs, outputs) return model def build_minimal_resnet_18(input_shape=(1800, 1), num_classes=10, dropout_rate=0.0): inputs = tf.keras.Input(shape=input_shape) reshaped_inputs = tf.keras.layers.Reshape((300, 6))(inputs) # Reshape to (300, 6) # Initial Convolution x = layers.Conv1D(32, kernel_size=7, strides=2, padding='same', use_bias=False)(reshaped_inputs) x = layers.BatchNormalization()(x) x = layers.Activation('relu')(x) x = layers.MaxPooling1D(pool_size=3, strides=2, padding='same')(x) # Simplified Residual blocks with increased filters x = simplified_residual_block(x, 32, dropout_rate=dropout_rate) # Additional block # x = simplified_residual_block(x, 32, dropout_rate=dropout_rate) x = simplified_residual_block(x, 64, stride=2, dropout_rate=dropout_rate) # x = simplified_residual_block(x, 64, dropout_rate=dropout_rate) # Additional block # x = simplified_residual_block(x, 64, dropout_rate=dropout_rate) # x = simplified_residual_block(x, 128, stride=2, dropout_rate=dropout_rate) # x = simplified_residual_block(x, 128, dropout_rate=dropout_rate) # Additional block # x = simplified_residual_block(x, 128, dropout_rate=dropout_rate) # x = simplified_residual_block(x, 256, stride=2, dropout_rate=dropout_rate) # x = simplified_residual_block(x, 256, dropout_rate=dropout_rate) # Additional block # Global Average Pooling and Output x = layers.GlobalAveragePooling1D()(x) outputs = layers.Dense(num_classes, activation='softmax', kernel_regularizer=reg)(x) model = models.Model(inputs, outputs) return model # Build and summarize the smaller model model = build_minimal_resnet_18(input_shape=(1800, 1), num_classes=10, dropout_rate=0.5) # model.summary() # this controls the learning rate opt = Adam(learning_rate=LEARNING_RATE, beta_1=0.9, beta_2=0.999) callbacks.append(BatchLoggerCallback(BATCH_SIZE, train_sample_count, epochs=EPOCHS, ensure_determinism=ENSURE_DETERMINISM)) # train the neural network model.compile(loss='categorical_crossentropy', optimizer=opt, metrics=['accuracy']) model.fit(train_dataset, epochs=EPOCHS, validation_data=validation_dataset, verbose=2, callbacks=callbacks) # Use this flag to disable per-channel quantization for a model. # This can reduce RAM usage for convolutional models, but may have # an impact on accuracy. disable_per_channel_quantization = False