问题
I am trying to apply Logistic Regression in Python using statsmodel.api.Logit. I am running into the error ValueError: array must not contain infs or NaNs.
When I am executing with:
data['intercept'] = 1.0
train_cols = data.columns[1:]
logit = sm.Logit(data['admit'], data[train_cols])
result = logit.fit(start_params=None, method='bfgs', maxiter=20, full_output=1, disp=1, callback=None)
The data contains more than 15000 columns and 2000 rows. which data['admit'] is the target value and data[train_cols] is the list of features. Can anyone please give me some hints to fix this problem?
回答1:
By default, Logit does not check your data for un-processable infinitities (np.inf) or NaNs (np.nan). In pandas, the latter normally signifies a missing entry.
To ignore rows with missing data and proceed with the rest, use missing='drop' like so:
sm.Logit(data['admit'], data[train_cols], missing='drop')
See the Logit docs for other options.
If you do not expect your data to contain any missing entries or infinities, perhaps you loaded it incorrectly. Look at data[data.isnull()] to see where the problem is. (N.B. Read this to see how to make infs register as null.)
来源：https://stackoverflow.com/questions/19223408/statsmodel-api-logit-valueerror-array-must-not-contain-infs-or-nans