Verification of dtypes of columns of X_row* is same that self.X #300
Conversation
…t problems in the prediction step.
salmuz
changed the title
| @@ -1791,3 +1796,25 @@ def get_xgboost_preds_df(xgbmodel, X_row, pos_label=1): | |||
| 0, "pred_proba" | |||
| ] | |||
| return xgboost_preds_df | |||
|
|
|||
|
|
|||
| def check_dtype_of( | |||
There was a problem hiding this comment.
could you add some tests for this? How flexible is it? (e.g. will it break over float32 vs float64? int vs float? etc)
| @@ -50,6 +50,7 @@ | |||
|
|
|||
|
|
|||
| from .explainer_methods import * | |||
| from .explainer_methods import check_dtype_of | |||
There was a problem hiding this comment.
you can add check_dtype_of to the __all__ at the start in explainer_methods.py then it is covered by the import * (generally import * is frowned upon, but it's okay as long as you define a restrictive __all__)
| @@ -241,7 +242,9 @@ def __init__( | |||
| col for col in self.regular_cols if not is_numeric_dtype(self.X[col]) | |||
| ] | |||
| self.categorical_dict = { | |||
| col: sorted(self.X[col].unique().tolist()) for col in self.categorical_cols | |||
| col: sorted( | |||
| v for v in self.X[col].unique().tolist() if not pd.isna(v) | |||
There was a problem hiding this comment.
not an expert on lightgbm, but wouldn't there be usecases where na would be a category? Or is that handled differently? How about by catboost or other libraries?
| df_target is not None and | ||
| not df_target[features].dtypes.eq(df_origin[features].dtypes).all() | ||
| ): | ||
| df_target[features] = df_target[features].astype( |
There was a problem hiding this comment.
in general not a fan of these functions that modify in place. Could you rewrite it such that it returns the transformed df instead? Then maybe call it adjust_dtypes_to_match_df(...) or something?
Calling something check_dtype_of when it actually modifies one of the arguments is confusing.
|
cool, thanks! tests are passing, but please have a look at my comments and see if you can add a few test cases for this new function... |
|
Hello, I will do the requested changes as soon as possible (the next week). Thanks |
Hello, I want to contribute by fixing two different bugs that are related to the usage of Ligthgbm.