Explain Predictions not working

[xnmp]

So I have a model called gbm. gbm.predict_proba(X) works, and explain_weights(gbm) works. But when I try to write

explain_prediction(gbm, X.iloc[1])

I get ValueError: DataFrame.dtypes for data must be int, float or bool. Did not expect that data types in fields (every field).

If I instead write

explain_prediction(gbm, X.iloc[1].values)

I get an answer, but all the features are label x1,x2,x3, etc.

Any idea what's going on?

[lopuhin]

hi @xnmp ! Can you please give the original traceback you get? I tried to reproduce the issue, but didn't get an error - probably you are doing something differently:

import pandas as pd
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.datasets import load_iris

data = load_iris()
data.keys()

X = pd.DataFrame(data['data'], columns=data['feature_names'])

gbm = GradientBoostingClassifier()
gbm.fit(X, data['target'])

eli5.show_weights(gbm)
eli5.show_prediction(gbm, X.iloc[1])

[lextoumbourou]

I'm having the same issue with a LightGBM. I found a simple workaround, which is to specify the columns using the feature_names arg:

explain_prediction(gbm, X.iloc[1].values, feature_names=list(X.columns))

[agnesvanbelle]

I had the same problem and above did not solve it for me. What worked for me was using X.iloc[[1]] instead of X.iloc[1]. The latter form automatically converts a row to a series, which converts the datatypes in the row to "object" type if they differ.