pytorch-forecasting VS pytorch-lightning

Source: Conversation with Bing, 4/5/2023 (1) jdb78/pytorch-forecasting: Time series forecasting with PyTorch - GitHub. https://github.com/jdb78/pytorch-forecasting. (2) Time Series Prediction with LSTM Using PyTorch - Colaboratory. https://colab.research.google.com/github/dlmacedo/starter-academic/blob/master/content/courses/deeplearning/notebooks/pytorch/Time_Series_Prediction_with_LSTM_Using_PyTorch.ipynb. (3) time-series-classification · GitHub Topics · GitHub. https://github.com/topics/time-series-classification. (4) PyTorch: Dataloader for time series task - Stack Overflow. https://stackoverflow.com/questions/57893415/pytorch-dataloader-for-time-series-task.

Temporal fusion transformer https://github.com/jdb78/pytorch-forecasting

Pytorch-forecasting

To name a few: https://github.com/jdb78/pytorch-forecasting, https://github.com/unit8co/darts, https://github.com/Nixtla/neuralforecast

I'm a data trainee at this organisation. I wrote my master thesis about using an event clustering mechanism to enrich an existing dataset to improve short-term demand predictions, using Pytorch Forecasting using the temporal fusion transformer component, and LightGBM (and compare the models with and w/o the event feature, so 4 runs in total).

Darts is a pretty nice one. I've recently been using pytorch-forecasting for larger models like the Temporal Fusion Transformer. https://github.com/jdb78/pytorch-forecasting

https://see.stanford.edu/Course/CS229 https://lightning.ai/ https://www.youtube.com/watch?v=00s9ireCnCw&t=57s https://towardsdatascience.com/

I've been wondering on what is the recommended method of saving logits/activations using PyTorch Lightning. I've looked at Callbacks, Loggers and ModelHooks but none of the use-cases seem to be for this kind of activity (even if I were to create my own custom variants of each utility). The ModelCheckpoint Callback in its utility makes me feel like custom Callbacks would be the way to go but I'm not quite sure. This closed GitHub issue does address my issue to some extent.

Our codebase for the diffusion models builds heavily on OpenAI's ADM codebase , lucidrains, Stable Diffusion, Lightning and Hugging Face. Thanks for open-sourcing!

For lightweight use, pytorch-lightning is too heavy, and its source code will be very difficult for beginners to read, at least for me.

For the A100 GPU, theoretical performance is the same for FP16/BF16 and both rely on the same number of bits, meaning memory should be the same. However since it's quite newly added to PyTorch, performance seems to still be dependent on underlying operators used (pytorch lightning debugging in progress here).