releaunifreiburg / welltunedsimplenets Goto Github PK

Hi,
looking to run a comparison of your model vs a few of the current transformer models on some tabular data that I have
In the installation instructions, after the gxx and gcc install, it states to install the requirements.txt, but I don't see one in the directory? Am I missing something?
Note: I am on windows and was unable to install gxx and gcc, unsure if that has an effect on this
thanks,
Jonathan

Why do you do cat requirements.txt | xargs -n 1 -L 1 pip install instead of pip install -r requirements.txt?

Hi author, thanks for sharing the code. I'm wondering how data augmentation strategies like mix-up, cut-out, cut-mix, etc., can be applied to tabular data (I understand they are usually applied to images though). Please advise, many thanks.

Hi there,

I stumbled across your project and paper and it seemed highly interesting. Unfortunately, I cannot successfully execute your project with my dataset. The following error occurs consistently

Traceback (most recent call last):
  File "cocktails/main_experiment.py", line 399, in <module>
    train_predictions = fitted_pipeline.predict(X_train)
AttributeError: 'NoneType' object has no attribute 'predict'

I have checked, and double-checked the dataset but cannot find "the problem". All values are fine, properly scaled etc. etc.
The onlything thats maybe remarkable is: The dataside is a bit wide: 80000 x 650 bu nothing too out of bounds.

Do you have an idea what I should check?
Thanks for your support

Kind regards

Hi!

Loved the paper. Wondering how one could adapt your repo to run a refit experiment on a custom dataset, rather than those on openML?

I get problems, that were solved when I replaced this if with an elif. I only understand the code on a very surface level, though. Not sure this is the right fix for these problems. It seems like 2 things happen: i) you assign to the same node twice (imputer) ii) the one-hot encoder is not supported for non-categorical features.
https://github.com/releaunifreiburg/WellTunedSimpleNets/blob/be5b3f3ccd3a5000b5108c19dc65c05357428fbc/utilities.py#L304

I ran into your paper not too long ago and found it pretty interesting, thanks for sharing the code.

I'm looking to run this on my own dataset and I'm a bit confused as to the cash_cocktail option in main_experiment.py. My impression is that this automatically turns on all the options to search for HPO but when I run the code I get output that looks like this:

{'task_id': 233088, 'wall_time': 9000, 'func_eval_time': 1000, 'epochs': 105, 'seed': 11, 'tmp_dir': './runs/autoPyTorch_cocktails', 'output_dir': './runs/autoPyTorch_cocktails', 'nr_workers': 6, 'nr_threads': 1, 'cash_cocktail': True, 'use_swa': [False], 'use_se': [False], 'use_lookahead': [False], 'use_weight_decay': [False], 'use_batch_normalization': [False], 'use_skip_connection': [False], 'use_dropout': [False], 'mb_choice': 'none', 'augmentation': 'standard'}
{'task_id': 233088, 'wall_time': 9000, 'func_eval_time': 1000, 'epochs': 105, 'seed': 11, 'tmp_dir': './runs/autoPyTorch_cocktails', 'output_dir': './runs/autoPyTorch_cocktails', 'nr_workers': 6, 'nr_threads': 1, 'cash_cocktail': True, 'use_swa': [False], 'use_se': [False], 'use_lookahead': [False], 'use_weight_decay': [False], 'use_batch_normalization': [False], 'use_skip_connection': [False], 'use_dropout': [False], 'mb_choice': 'none', 'augmentation': 'standard'}

It looks like the options aren't being used? e.g. 'use_swa': [False], 'use_se': [False], 'use_lookahead': [False],...

I guess to rephrase the question: if one were to use your code on their private dataset, what options do you need to pass in to ensure that you are doing the full HPO? is it just the cash_cocktail flag?

Hi,

thanks for sharing the great work.

It seems that it is not compatible with the current version of AutoPytorch. I made following changes in utilities.py to make it work. Can you make sure my changes below are reasonable?

attached is utilities.py diff file in text

utilities.py.diff.txt

For me the example breaks, if I do not fix smac to smac==1.1.1