udellgroup / controlburn Goto Github PK

Hello,

Is there a conda version of controlburn?

Is it under development or already done?

I get the following error when trying to fir a CB classifier model on my data. I get the error after preprocessing (to impute nans, remove strings etc'). (I also get a different error if I try using CB as part of a pipeline).

X2 = preprocessor.fit_transform(X)
cb = ControlBurnClassifier
cb.fit(X2,y)


AttributeError                            Traceback (most recent call last)
/tmp/ipykernel_17181/3413679809.py in <module>
----> 2 cb.fit(X2,y)

~/.local/lib/python3.7/site-packages/ControlBurn/ControlBurnModel.py in fit(self, X, y, costs, groups, sketching)
    700 
    701         """
--> 702         self.build_forest_method(X,y)
    703         self.solve_lasso( costs = costs, groups = groups, sketching = sketching)
    704         if len(self.features_selected_) == 0:

~/.local/lib/python3.7/site-packages/ControlBurn/ControlBurnModel.py in bagboost_forest(self, X, y)
    276         y = pd.Series(y)
    277 
--> 278         X = X.reset_index().drop('index',axis = 1)
    279         y.index = X.index
    280 

**AttributeError: 'numpy.ndarray' object has no attribute 'reset_index'**

Are there plans to update the package to allow it's use in scikit-learn pipelines?
(It currently lacks support for all the required methods. e.g. it can't be used in a pipeline with SelectFromModel(cb) for example.

Hello,

Can you please let me know if there were special computing requirements you utilized while running your experiments? like gpu's, memory, etc?
It seems I'm running out of space at the cell that uses ray library.

Would appreciate any suggestion or advice.

Specifically, this is what I receive, when running this line ray.get(result_ids)

RaySystemError: System error: __init__() missing 1 required positional argument: 'msg'
traceback: Traceback (most recent call last):
  File "env/lib/python3.8/site-packages/ray/serialization.py", line 254, in deserialize_objects
    obj = self._deserialize_object(data, metadata, object_ref)
  File "env/lib/python3.8/site-packages/ray/serialization.py", line 213, in _deserialize_object
    return RayError.from_bytes(obj)
  File "env/lib/python3.8/site-packages/ray/exceptions.py", line 28, in from_bytes
    return pickle.loads(ray_exception.serialized_exception)
TypeError: __init__() missing 1 required positional argument: 'msg'

Hi, thank you for this library! Question regarding feature importance amongst 1000s of features. Do you think that this library will work as well as potentially others such as SHAP on 1000s of features for selection? And advice would be appreciated. Thank you.

I read through the paper of controlburn and it is really inspiring. However I have one question about the experiments. Since controlburn is motivated by LASSO, why it is not compared with LASSO in the experiments? Why don't we use LASSO directly instead?

Hi, I tried to run controlburn.classifier on a dataset that has 62301 rows, and 1208 features, but it only outputs 2 features when I set alpha = 0.05. When I tried smaller alpha values like 0.005 or 0.0005, it raised an error "SolverError: Solver 'ECOS_BB' failed. Try another solver, or solve with verbose=True for more information.".

Thank you for reviewing the issue!