Practice convolutional net for MNIST.
https://medium.com/@nutanbhogendrasharma/pytorch-convolutional-neural-network-with-mnist-dataset-4e8a4265e118 Using PyTorch builtins like dataloader, dataset, sequential conv layers (conv, ReLu, Pool)(which is new to me).
Looks like this is overfitting, or an extremely network architecture to solve this problem, since accuracy is 100% after 2400 samples.
https://wandb.ai/lindlaaa/mnist-conv/reports/Untitled-Report--VmlldzoyOTM1Nzgy