Create a new version of the MNIST dataset that does not contain the numbers “9”or “5”, and then train one of the auto encoders on this dataset. Then run the autoencoder on the test dataset, and record...

Create a new version of the MNIST dataset that does not contain the numbers “9”or “5”, and then train one of the auto encoders on this dataset. Then run the autoencoder on the test dataset, and record the average error (MSE) for each of theten classes. Do you see any patterns in the results, and is it able to identify “9” and“5” as outliers? (This has to be acheived only using Pytorch)
This is for reference-https://github.com/EdwardRaff/Inside-Deep-Learning/blob/main/Chapter_7.ipynb