I am trying to train the model both in coursera and on colab, I was able to pass assignment 1, but I can’t train the model it gives me this error,

InvalidArgumentError Traceback (most recent call last)

in

6 num_batches = 0

7 for x in tqdm(train_dist_dataset):

----> 8 total_loss += distributed_train_step(x)

9 num_batches += 1

10 train_loss = total_loss / num_batches

InvalidArgumentError: Incompatible shapes: [64] vs. [48] [Op:AddV2]

Please dont publish solutions code here.

The error tells you there is a shape mismatch in the distributed_train_step(x) and probably traced back in the train_step there must be a misshape somewhere there or even further up you need to check (maybe predictions and labels shapes are different). But surprising is the fact that you already passing the assignment!

Thank you i was able to solve it, I used a wrong loss function, i am trying to upload my saved model to the coursera platform, but it isn’t working what can be the solution?

You mean you are tyring to upload a notebook to coursera environment and you can’t?

In the Practice Programming Assignment, where we are asked to upload our model, i tried submitting the zip file and it didn’t upload, i tried it part by part, first sending saved_model.pb then the variables but it didnt work either, i tried uploading the notebook and it worked, but the grader gave me zero because my file couldn’t be unzipped, how do i go about it?

I see, I just uploaded it myself and passed 100, so maybe you have altered the zipfile? I say you rerun and reproduce the zipfile once again and try again.

What was the size of your compressed Zip folder? my zip folder had a size of 86.5mb, I ran the code in colab as instructed and downloaded the compressed zip folder to upload I don’t know what could be wrong

It has finally been uploaded and I have received a grade, thank you so much for your help, it is highly appreciated.

1 Like