search

timeseriesAI / tsai

Time series Timeseries Deep Learning Machine Learning Python Pytorch fastai | State-of-the-art Deep Learning library for Time Series and Sequences in Pytorch / fastai
https://timeseriesai.github.io/tsai/
Apache License 2.0
5.15k stars 643 forks source link

Allow for parameterizing block_szs in XResNet1dPlus #730

Closed talesa closed 1 year ago

talesa commented 1 year ago

Minor change to allow to parameterize block_szs in XResNet1dPlus. Needed it to reproduce some results elegantly.

Also, fixing a small bug with custom_head argument for some of the models. The same solution as per other models, e.g., https://github.com/timeseriesAI/tsai/blob/00593d5ef3e0b3fc82bc7bd54fd325bab8d615d2/tsai/models/ResNetPlus.py#L65-L67

cc @oguiza

review-notebook-app[bot] commented 1 year ago

Check out this pull request on  ReviewNB

See visual diffs & provide feedback on Jupyter Notebooks.

Powered by ReviewNB

oguiza commented 1 year ago

Thanks a lot, @talesa for putting together this PR! The PR is now merged, so you are a new tsai contributor!

I have a question and a request. You mention "Needed it to reproduce some results elegantly." I'm curious to know what results you are trying to reproduce. AFAIK XResNet1dPlus is a model I adapted from fastai's vision xresnet, and it's only available in tsai. I'm not aware that anyone has published any results with it. In the future, it'd be good to keep PRs focused on a single enhancement/ bug fix. The changes you made make sense, but you are mixing an enhancement in one model with a bug fix in 2 different models.

Thanks again and keep up the great work!!