Sang Michael Xie

At least as of v1.2.2, we do the augmentation: https://github.com/p-lambda/wilds/blob/6d96cff360018bdb7c0863ba3976f7fa646aaaab/examples/transforms.py#L99 And the normalization occurs during preprocessing: https://github.com/p-lambda/wilds/blob/472677590de351857197a9bf24958838c39c272b/dataset_preprocessing/poverty/convert_poverty_to_npy.py#L35 However I'm not sure about how the v2 codepath works. @teetone @kohpangwei

Sorry for the late response! We haven't run that before, but it would be interesting. To do it, you just have to modify the list here: https://github.com/p-lambda/incontext-learning/blob/fdf346bc233fd399f2a97fdf9cc44eccc08c508a/generate_data.py#L286 The structure of...

Could you provide some more details? What were the per-task results that you got? Did you use the quality filter that filters for length, numeric ratio, etc? Did you preprocess...

Ah, just found a typo that was introduced when fixing the `domain_to_idxs` issue earlier: https://github.com/p-lambda/dsir/blob/cb7b6c61cd14fe7b2e6bc0774f805b6b6f94d235/experimental/data_selection/dsir_pipeline.py#L224 Could you try running the resampling step again?

> Ah, just found a typo that was introduced when fixing the `domain_to_idxs` issue earlier: > > https://github.com/p-lambda/dsir/blob/cb7b6c61cd14fe7b2e6bc0774f805b6b6f94d235/experimental/data_selection/dsir_pipeline.py#L224 > > Could you try running the resampling step again? Did you...

When you print `total_idx` in your code, is the number matching 1745766302? > Thank you for clarifying my confusion. Are you saying that you use the token distributions to compute...

The code doesn't work for newer versions of TF right now. It does use GPU. On Thu, Apr 4, 2019 at 6:36 PM Kuan Lu(Frank) wrote: > Okay, I will...

You should check your LD_LIBRARY_PATH and associated environment variables that cuda is correctly linked, and that tensorflow-gpu is installed (without tensorflow also installed). For some of the methods (like label...

Sorry this is late - How much greater than 1 is it? It might just be a numerical issue that could be resolved by renormalizing.

You're right that in principle we could use all the unlabeled data we have. However for our purposes we look at the transductive setting where unlabeled data = test data....