Google colab - Feature selection not working

[AlizeeL]

This is my setting: dataframe dataset, numerical values, Target is binary classification, I am trying to do feature selectection.
automl = AutoML(
mode = 'Compete',
eval_metric = 'f1',
validation_strategy = {"validation_type": "custom"},
results_path=folder+'automl_featsel2_'+subject_val,
explain_level = 1,
golden_features = False,
algorithms = ['Xgboost'],
features_selection = True,
stack_models = False,
hill_climbing_steps = 0,
top_models_to_improve = 5,
train_ensemble = False,
start_random_models = 1,
kmeans_features = False,
random_state = 42
)

Hello, I get the following warning when I fit:

log_loss_eps() got an unexpected keyword argument 'response_method'
Problem during computing permutation importance. Skipping ...
'module' object is not callable

Skip features_selection because no parameters were generated.

[pplonski]

thanks @AlizeeL for reporting, it looks like a bug.

May I ask why are you using Colab? do you need a lot of computational power?

[AlizeeL]

Some of the datasets I am using can be quite big so yes. Using Colab is a side of my research on accessibility to such tools to non-expert users.

[pplonski]

@AlizeeL thanks for response, we are working on notebook with UI for code generation, that is designed for non-experts users. It is called MLJAR Studio, available as desktop app on our website https://mljar.com/ It is in early development phase, but csv data loading and AutoML training is working. I hope you will find it interesting.

[AlizeeL]

Thanks @pplonski , it does look promising.

Do you know if my type of issue might get solved in the near future? I just need to know in case I have to work on a machine instead of Colab.

[pplonski]

Thank you. I'm adding @Bocianski to disscussion about plans for fix. For sure, it will help us a lot, if you could provide full code and data for reproduction.

[AlizeeL]

Here's my code. There's a link at the top to a dataset.
It's a reduced version of my dataset to avoid long computational time. It doesn't change the issues/output.

Let me know if there's any problems with the links, I can send code/data by email if that's the case.
Thanks :)

[Reese-Martin]

i am seeing a similar error while running MLJar on an azure VM, do y'all know why this may be happening?

this is the specific error
"log_loss_eps() got an unexpected keyword argument 'response_method'
Problem during computing permutation importance. Skipping ..."