Continuing the discussion from Programming Assignment: Multiple Output Models using Keras Functional API:
All my tests ran okay, with outputs still not passing the test
Course Q&A
TensorFlow: Advanced Techniques Specialization
Custom Models, Layers and Loss Functions with TF
What’s your problem? I read the topic you are linking and can’t understand if you are having the same problem or maybe a different issue.
But the other mentors have recommended resetting the assignemet.
Actually all my tests ran okay, all outputs matched with desired results, all graphs perfect.
Unfortunately the assignment on submission gave 3/10 and failed.
I passed all other tests of that module ( week 2 to week 5) with 100 percentage marks. Only this assignment is causing issue.
can i send you my file?, it’s perfectly okay
Yes send it to me! I will revise it in some hours this night.
my file is attached with this mail
[CODE REMOVED FROM MODERATOR]
my file is attached with this mail
1 Like
Hi @Anindya_kumar_roy ,
Don’t ask me why, but I think that the problem can be the URI of the files:
Try to use:
URI = ‘./winequality-red.csv’
URI = ‘./winequality-white.csv’
Instead of the full path, maybe the problem is that the grader can’t acces the same directory than the lab.
I used your code and get 3/10, but just changed ythis url and got a 10/10.
hop it helps. If this is the solution, please let-me know, maybe we can indicate it in tne notebooks for future students.
And mark this reply as a solution. It can help other students too!
Yes, it is solved, program was right, url was problematic 
I am having an issue in the last loss parameter of week 3 of Generative Deep Learning with TensorFlow. I am sending my program file, kindly check and guide me.
From,
Anindya
SHARING CODE SOLUTION PUBLICLY IS AGAINST HONOR OF CONDUCT
You can include a description of the issue at best, but sharing code publicly is not a practice allowed here.
Okay.
I am having an issue in the last loss parameter of week 3 (variational auto encoders on anime faces) of Generative Deep Learning with TensorFlow. All other parts are okay, and the programs ran efficiently till the last part.
During the run of last training loop an error occurred, shape/size mismatch, indicating the loss step, any help in this regard is highly appreciated. Except that last step all other parts are okay and the program ran smoothly till that point.
From,
Anindya
Okay.
I am having an issue in the last loss parameter of week 3 (variational autoencoders on anime faces) of Generative Deep Learning with TensorFlow. All other parts are okay, and the programs ran efficiently till the last part.
During the run of the last training loop an error occurred, shape/size mismatch, indicating the loss step, any help in this regard is highly appreciated. Except that last step all other parts were okay and the program ran smoothly till that point.
the error:
InvalidArgumentError: Incompatible shapes: [24576000] vs. [8192000] [Op:Mul]
From,
Anindya
Hi @Anindya_kumar_roy, possible places where the the error could be:
- Kullback–Leibler Divergence
- vae_model(encoder, decoder, input_shape)
- How you calculate loss with gradient_tape
There’s where I would start checking initially!
Yes, in encoder and decoder layer, the decoder output I mistakenly set at 1. That’s where the error was. Encoder layer - I/p 32,64,128. Decoder was 128,64,32 and final filter was 1, it should be 3.
As image 64,64,3. Input and output size mismatch.
I changed it, now it’s okay.
One more suggestion, due to long time I set the iterations/epochs = 50. Not 100.
Can there be issue?
From,
Anindya
1 Like
Try with 50 and if not successful you can increase gradually the number of epochs.
Yes 50 I am doing. On process.
Thanks for the help.
From,
Anindya
1 Like