In the Autoencoder assignment, I successfully created the model, and it is learning correctly, but the loss is not going below 0.01, and accuracy is not increasing than 0.6.
Someone, please guide me why it is not improving.
I already, use different architecture and change number of layers but no improvement in the result to pass the assignment.
Below is the notebook.
Copy_of_C4W2_Assignment.ipynb (127.6 KB)
Waqar please shift the post to the correct column. select the correct course, then week. You have selected the correct specialisation after that you didn’t select the correct course and week
Kindly DM your notebook, to have a look at where or how your model algorithm is not getting the desired result. Click on my name, then message.
I am still not able to increase the accuracy. it should be greater than 0.6 to pass the grade but its not increasing.
Can someone guide me through the problem?
I just went through your notebook.
In the load and prepare dataset section, although you have written the correct code lines but added hashtag for the right one and using the wrong code line for execution. please check the below statement to do correction for these two code lines.
shuffle and batch the dataset ==INCORRECT CODE (CORRECT CODE WRITTEN WITH HASHTAG)
batch the dataset ===INCORRECT CODE (CORRECT CODE WRITTEN WITH HASHTAG)
ANOTHER MAJOR MISTAKE IN YOUR NOTEBOOK IS IN YOUR BUILD THE MODEL GRADER CELL. WHILE YOU REFERRING THE UNGRADED LABS DOES SHOW TO DEFINE EACH LAYER, BUT IN THE ASSIGNMENT DO NOT DEFINE EACH LAYER SEPARATELY.
REMEMBER YOU ONLY NEED TO CREATE AN AUTOENCODER ALGORITHM WHICH INCLUDES INPUT LAYER, CONV2D LAYER AND UPSAMPLING2D LAYERS, AND YOUR OUTPUT. BETWEEN YOUR MAXPOOLING AND UPSAMPLING LAYER IS WHERE YOUR BOTTLE NECK AND ENCODER VISUALIZATION LAYER COMES. ONCE AT THE END YOU HAVE RECALLED OUTPUT LAYER WITH THE CORRECT FILTER, MODEL NEED TO BE DEFINE BY THIS LAYER model = tf.keras.Model(inputs, outputs)
REFER THIS VIDEO AGAIN AND INCLUDE ONLY THE LAYERS MENTIONED IN THE VIDEO BY LAWRENCE.
ALSO MAKE SURE YOUR UNIT IS CORRECT WITH RESPECT TO CIFAR-10 CODER. YOUR INPUT LAYER IS INCORRECT W.R.T. TO SHAPE. REFER THIS PAGE TO MAKE CORRECTIONS
CHECK YOU ACTIVATION FOR encoder_visualization, ARE YOU USING THE CORRECT ACTIVATION FOR THIS LAYER, REFER THE WEEK 2 VIDEO AGAIN.
THIS SECTION IS NOT REQUIRED. REMEMBER YOUR DECODER LAYER IS CREATING THE AUTOENCODER LAYER. SO YOU DO NOT NEED TO DEFINE SEPARATELY AN AUTOENCODER MODEL.
‘’‘Builds the entire autoencoder model.’‘’
ALSO YOUR CONV4 LAYER SHOULD BE CALLED OUTPUT AND WHOSE FILTER IS CORRECT IN YOUR MODEL. REFER THE CIFAR-10 DOCUMENT AGAIN AND DO THE CORRECTION.
IF YOU STILL FIND DIFFICULT TO DO THE CORRECTION, LET ME KNOW. TRY ONCE YOURSELF.