- Using cuda() to move data to GPU. Original sample runs on CPU! GPU is far slower (double the time) than CPU
- RNN from scratch New to pyTorch, as usual, wanted to learn by doing things from scratch. Below are a few blogs that got me going.
Number 1, really from scratch https://towardsdatascience.com/understanding-pytorch-with-an-example-a-step-by-step-tutorial-81fc5f8c4e8e?#3a3f
ImageRNN but it is using nn.RNN, I wanted even lower.. 🙂 https://medium.com/dair-ai/building-rnns-is-fun-with-pytorch-and-google-colab-3903ea9a3a79
PyTorch – Tutorial that set me thinking..but this is text classification not using Dataloader and batch processing, as I consider batching is tricky one to get going, so, I wanted to indulge on it 😉
https://pytorch.org/tutorials/intermediate/char_rnn_classification_tutorial.html https://pytorch.org/tutorials/beginner/former_torchies/nnft_tutorial.html
More details here https://ojaslabs.com/pytorch-rnn-from-scratch