deep transfer cnn completed

BackEnd/deep_transfer_cnn_classifier.py

@@ -130,6 +130,124 @@ class DeepCNNClassifier:
     print("=========== Data Pre-Processing Completed ===========")
     return (train_data_ , test_data_)
   
+   #----------------------------------------------------WILL_USE_BY_ANOTHER_FUNCTION-----------------------------------------------------------------------------
+  def create_model(self):
+    import tensorflow as tf
+    import tensorflow_hub as hub
+    from tensorflow.keras import layers
+
+    #Download the pretrained model and save it as a keras layer
+    feature_extractor_layer = hub.KerasLayer(self.transfer_model_url,
+                                            trainable = False, #freeze the already learned patterns 
+                                            name = "feature_extraction_layer",
+                                            input_shape = (self.image_height, self.image_width, self.num_color_channels)) 
+    self.activation_func = 'sigmoid'
+    if(self.number_of_classes > 2):
+      self.activation_func = 'softmax'
+
+
+    model  =  tf.keras.Sequential([
+      feature_extractor_layer,
+      layers.Dense(self.number_of_classes , activation= self.activation_func , name="output_layer")
+    ])
+    print("=========== Model is Created ==========")
+    return model
+  
+  
+  #---------------------------------------------------SHOULD_CALL_BY_USER------------------------------------------------------------------------------
+  def start_training(self , num_epochs = 5):
+    import tensorflow as tf
+    train_data_ , test_data_ = self.preprocess_dataset()
+    cnn_model =  self.create_model()
+
+    self.loss_function = 'binary_crossentropy'
+    if(self.number_of_classes > 2):
+      self.loss_function = 'categorical_crossentropy'
+
+    cnn_model.compile(loss=self.loss_function,
+                     optimizer = tf.keras.optimizers.Adam(),
+                     metrics=["accuracy"])
+    print("=========== Model Compiled SuccessFully ==========")
+
+    model_hist = cnn_model.fit(train_data_,
+                               epochs=num_epochs,
+                               steps_per_epoch=len(train_data_),
+                               validation_data = test_data_,
+                               validation_steps = len(test_data_))
+    print("=========== Model Trained SuccessFully ==========")
+    self.trained_model = cnn_model
+    self.model_train_history = model_hist
+    return cnn_model
+  
+  #---------------------------------------------------SHOULD_CALL_BY_USER------------------------------------------------------------------------------
+  def save_model(self, save_to_path = '/content'):
+    model_name = str(self.model_name)+".h5"
+    model = self.trained_model
+    model.save(model_name)
+    print("=========== Model Saved to -->  {}".format(model_name))
+
+  #---------------------------------------------------SHOULD_CALL_BY_USER / Function------------------------------------------------------------------------------
+  def plot_loss_curves(self, ):
+    import matplotlib.pyplot as plt
+
+    history = self.model_train_history
+    loss = history.history["loss"]
+    val_loss = history.history["val_loss"]
+
+    accuracy = history.history["accuracy"]
+    val_accuracy = history.history["val_accuracy"]
+
+    #get the number of epochs that we run for
+    epochs =  range(len(history.history["loss"]))
+
+    #Plot the lost
+    plt.plot(epochs , loss , label="Training Loss")
+    plt.plot(epochs , val_loss , label="Validation Loss")
+    plt.title("Loss")
+    plt.xlabel("Epochs")
+    plt.legend()
+
+    #Plot the accuracy
+    plt.figure()
+    plt.plot(epochs , accuracy , label="Training accuracy")
+    plt.plot(epochs , val_accuracy , label="Validation accuracy")
+    plt.title("accuracy")
+    plt.xlabel("Epochs")
+    plt.legend()
+  
+  #----------------------------------------------------WILL_USE_BY_ANOTHER_FUNCTION-----------------------------------------------------------------------------
+  def load_and_prep_image(self, filename):
+    import tensorflow as tf
+
+    img = tf.io.read_file(filename)
+    img = tf.image.decode_image(img)
+    #Resiz the Image
+    img = tf.image.resize(img , size=[self.image_height , self.image_width])
+    #Re scale the image - get all values between 0 and 255
+    img = img/255.
+    return img
+
+  #---------------------------------------------------SHOULD_CALL_BY_USER------------------------------------------------------------------------------
+  def pred_and_plot(self, filename):
+    import tensorflow as tf
+    import matplotlib.pyplot as plt
+    import numpy as np
+
+    model = self.trained_model
+    class_names = self.class_names
+  
+    img = self.load_and_prep_image(filename)
+    
+    pred = model.predict(tf.expand_dims(img , axis=0))
+
+    pred_class =  class_names[np.argmax(tf.round(pred))]
+
+    #Plot the image and the predicted class
+    plt.imshow(img)
+    plt.title(f"Prediction :{pred_class}")
+    plt.axis(False)
+
+