Update test.py

test.py

@@ -1,57 +1,53 @@
 import tensorflow as tf
 from tensorflow.keras.models import Model
-from tensorflow.keras.layers import Input, Dense, Dropout, BatchNormalization
+from tensorflow.keras.layers import Input, Dense, Dropout, BatchNormalization, Activation
 from tensorflow.keras import regularizers
 from tensorflow.keras.optimizers import Adam
-from tensorflow.keras.callbacks import EarlyStopping, ModelCheckpoint
+from tensorflow.keras.callbacks import EarlyStopping, ModelCheckpoint, ReduceLROnPlateau
 
 with tf.device('/gpu:0'):
-    # this is the size of our encoded representations
     encoding_dim1 = 500
     encoding_dim2 = 200
-    
+
     lambda_act = 0.0001
     lambda_weight = 0.001
-    
-    # this is our input placeholder
+
     input_data = Input(shape=(num_in_neurons,))
-    
-    # first encoded representation of the input
-    encoded = Dense(encoding_dim1, activation='relu', 
-                    activity_regularizer=regularizers.l1(lambda_act),
-                    kernel_regularizer=regularizers.l2(lambda_weight))(input_data)
+
+    # Encoder
+    encoded = Dense(encoding_dim1, activity_regularizer=regularizers.l1(lambda_act), kernel_regularizer=regularizers.l2(lambda_weight), name='encoder1')(input_data)
     encoded = BatchNormalization()(encoded)
+    encoded = Activation('relu')(encoded)
     encoded = Dropout(0.5)(encoded)
-    
-    # second encoded representation of the input
-    encoded = Dense(encoding_dim2, activation='relu', 
-                    activity_regularizer=regularizers.l1(lambda_act),
-                    kernel_regularizer=regularizers.l2(lambda_weight))(encoded)
+
+    encoded = Dense(encoding_dim2, activity_regularizer=regularizers.l1(lambda_act), kernel_regularizer=regularizers.l2(lambda_weight), name='encoder2')(encoded)
     encoded = BatchNormalization()(encoded)
+    encoded = Activation('relu')(encoded)
     encoded = Dropout(0.5)(encoded)
-    
-    # first lossy reconstruction of the input
-    decoded = Dense(encoding_dim1, activation='relu')(encoded)
+
+    # Decoder
+    decoded = Dense(encoding_dim1, name='decoder1')(encoded)
     decoded = BatchNormalization()(decoded)
-    
-    # the final lossy reconstruction of the input
-    decoded = Dense(num_in_neurons, activation='sigmoid')(decoded)
-    
-    # this model maps an input to its reconstruction
+    decoded = Activation('relu')(decoded)
+
+    decoded = Dense(num_in_neurons, name='decoder2')(decoded)
+    decoded = Activation('sigmoid')(decoded)
+
     autoencoder = Model(inputs=input_data, outputs=decoded)
-    autoencoder.compile(optimizer=Adam(), loss='mse')
-    
-    # setup callbacks
+    autoencoder.compile(optimizer=Adam(learning_rate=0.001), loss='mse')
+
+    # Callbacks
     callbacks = [
-        EarlyStopping(monitor='val_loss', patience=5, verbose=1),
-        ModelCheckpoint('best_model.h5', monitor='val_loss', save_best_only=True, verbose=1)
+        EarlyStopping(monitor='val_loss', patience=10, verbose=1),
+        ModelCheckpoint('best_model.h5', monitor='val_loss', save_best_only=True, verbose=1),
+        ReduceLROnPlateau(monitor='val_loss', factor=0.1, patience=5, verbose=1)
     ]
-    
-    # training
+
+    # Training
     print('Training the autoencoder')
     autoencoder.fit(x_train_noisy, x_train,
                     epochs=50,
-                    batch_size=16,
+                    batch_size=16,  # Adjusted batch size
                     shuffle=True,
                     validation_data=(x_test_noisy, x_test),
                     callbacks=callbacks)
@@ -59,10 +55,5 @@ with tf.device('/gpu:0'):
     # Load best model
     autoencoder.load_weights('best_model.h5')
 
-    # Freeze the weights
+    # Freeze the weights for inference
     autoencoder.trainable = False
-
-    # Use the trained encoder for predictions
-    myencoder = Model(inputs=input_data, outputs=encoded)
-    ae_train = myencoder.predict(x_train)
-    ae_test = myencoder.predict(x_test)