fashion-mnist | A MNIST-like fashion product database Benchmark :point_down: | Machine Learning library

The error is caused by the number of samples in the dataset and the batch size.

In more detail, the training MNIST dataset includes 60,000 samples, your current batch_size is 128 and you will need 60000/128=468.75 loops to finish training on one epoch. So the problem comes from here, for 468 loops, your data will have 128 samples but the last loop just contains 60000 - 468*128 = 96 samples.

To solve this problem, I think you need to find the suitable batch_size and the number of neural in your model as well.

I think it should work for computing loss

Well, this is a good question if you need to know the inner computation within your model. Let me explain to you!

So firstly when you print the model variable you'll get this output:

Your tensorboard logdir is not logs/fit.. but there is the current date appended. Try using a logs/fit as log_dir and see if it's working.

EDIT

If you want to use tensorboard locally you have to send tensorboard logs to S3 and read from there. In order to do this you have to do what your third linked example does, so include sagemaker debugger:

from sagemaker.debugger import TensorBoardOutputConfig

tensorboard_output_config = TensorBoardOutputConfig( s3_output_path='s3://path/for/tensorboard/data/emission', container_local_output_path='/local/path/for/tensorboard/data/emission' )

then your tensorboard command will be something like:

AWS_REGION= AWS_LOG_LEVEL=3 tensorboard --logdir s3://path/for/tensorboard/data/emission

Alternatively if you want to use tensorboard in the notebook you have to do what the second linked example does, so simply install in a cell and run tensorboard with something like:

https:///proxy/6006/

you can try like this:

Well you can do a simple expand dims:

nn.CrossEntropyLoss and nn.MSELoss are completely different loss functions with fundamentally different rationale behind them.

nn.CrossEntropyLoss is a loss function for discrete labeling tasks. Therefore it expects as inputs a prediction of label probabilities and targets as ground-truth discrete labels: x shape is nxc (where c is the number of labels) and y is of shape n of type integer, each target takes values in the range {0,...,c-1}.

In contrast, nn.MSELoss is a loss function for regression tasks. Therefore it expects both predictions and targets to be of the same shape and data type. That is, if your prediction is of shape nxc the target should also be of shape nxc (and not just n as in the cross-entropy case).

If you are insisting on using MSE loss instead of cross entropy, you will need to convert the target integer labels you currently have (of shape n) into 1-hot vectors of shape nxc and only then compute the MSE loss between your predictions and the generated one-hot targets.