search

Wenlin-Chen / ADKF-IFT

Official PyTorch Implementation of ADKF-IFT from Meta-learning Adaptive Deep Kernel Gaussian Processes for Molecular Property Prediction (ICLR 2023)
https://openreview.net/forum?id=KXRSh0sdVTP
Other
10 stars 1 forks source link

Some README updates. #4

Closed AustinT closed 1 year ago

AustinT commented 1 year ago
  1. Change environment name to adkf-fsmol: it's more specific and won't conflict with existing FS-Mol environments.
  2. Add explicit instructions to create environment and download dataset (before it took a bit of searching)
  3. Replace /path/to/dataset with the actual path one gets from downloading the dataset (it was not completely clear what path to use)
AustinT commented 1 year ago

Ultimately I think it would be nice to include pre-trained model weights and link to those in the README, and replace things like {model_name} with the actual string that the user should use (it's not clear whether it should be "ADKF" or something like "FSMol_ADKTModel_gnn+ecfp+fc_2023-02-08_09-56-57").

@Wenlin-Chen maybe you could make an extra commit to this branch making these changes?

Wenlin-Chen commented 1 year ago

@AustinT I have updated the README file. Where do you think we should upload the model weights to (figshare or this repo)? The size of the model weight file is about 100MB.

AustinT commented 1 year ago

@Wenlin-Chen I see you changed the name of the environment in README.md but not in environment.yml, was that intentional?

AustinT commented 1 year ago

@Wenlin-Chen one potential issue with the README now is that the different code snippets are not runnable in isolation: they would need to run the line dataset_dir=... in the same shell, which may not be immediately obvious.

AustinT commented 1 year ago

@AustinT I have updated the README file. Where do you think we should upload the model weights to (figshare or this repo)? The size of the model weight file is about 100MB.

Can the 100MB size be reduced by zipping, or is this already zipped?

Wenlin-Chen commented 1 year ago

@Wenlin-Chen I see you changed the name of the environment in README.md but not in environment.yml, was that intentional?

I forgot to change the environment.yml file. I'll fix it now. I think it's better to use adkf-ift rather than adkf to avoid confusion because ADKF is actually one of the ablation models in our paper.

Wenlin-Chen commented 1 year ago

@Wenlin-Chen one potential issue with the README now is that the different code snippets are not runnable in isolation: they would need to run the line dataset_dir=... in the same shell, which may not be immediately obvious.

Ok, that's a good point. I'll change it back

Wenlin-Chen commented 1 year ago

@AustinT I have updated the README file. Where do you think we should upload the model weights to (figshare or this repo)? The size of the model weight file is about 100MB.

Can the 100MB size be reduced by zipping, or is this already zipped?

It's not zipped. I'll zip it and see how much the size can be reduced

Wenlin-Chen commented 1 year ago

@AustinT I have updated the README file. Where do you think we should upload the model weights to (figshare or this repo)? The size of the model weight file is about 100MB.

Can the 100MB size be reduced by zipping, or is this already zipped?

We have two weight files (classification & regression), so 200MB in total. With highest zip compression level, the zipped file is 178MB.

AustinT commented 1 year ago

@AustinT I have updated the README file. Where do you think we should upload the model weights to (figshare or this repo)? The size of the model weight file is about 100MB.

Can the 100MB size be reduced by zipping, or is this already zipped?

We have two weight files (classification & regression), so 200MB in total. With highest zip compression level, the zipped file is 178MB.

Let's put it on Figshare or Zenodo then? Both are good options I think.