How to print the feature importance list in CPP API

guotong1988 · July 26, 2018, 10:34am

https://github.com/dmlc/xgboost/blob/master/include/xgboost/c_api.h

THX!!!

guotong1988 · July 26, 2018, 10:20am

XGBoosterPredict has a param which says “4:output feature contributions to individual predictions”

guotong1988 · July 26, 2018, 10:27am

But I don’t know how to use it.

hcho3 · July 27, 2018, 10:36pm

By “feature importance,” do you want something equivalent to http://xgboost.readthedocs.io/en/latest/python/python_api.html#xgboost.Booster.get_fscore ?

guotong1988 · August 2, 2018, 9:25am

Yes！THX！in CPP API, how to …

hcho3 · August 2, 2018, 6:07pm

There is no feature equivalent to get_fscore in C API. You’ll have to dump the model with XGBoosterDumpModel() and manually parse Gain and Cover to compute feature importance. See

github.com

dmlc/xgboost/blob/34dc9155abb81620f745b31bc096c5b4c74faf1e/python-package/xgboost/core.py#L1346


    """Get feature importance of each feature.


    Parameters
    ----------
    fmap: str (optional)
       The name of feature map file
    """


    return self.get_score(fmap, importance_type='weight')


def get_score(self, fmap='', importance_type='weight'):
    """Get feature importance of each feature.
    Importance type can be defined as:
        'weight' - the number of times a feature is used to split the data across all trees.
        'gain' - the average gain across all splits the feature is used in.
        'cover' - the average coverage across all splits the feature is used in.
        'total_gain' - the total gain across all splits the feature is used in.
        'total_cover' - the total coverage across all splits the feature is used in.


    Parameters
    ----------

for the logic of computing feature importance.