I think the error is in old version. I successfully recreated the error with the old version of scikit I had, I then updated it and I am not getting the error anymore, thoughts?

I also confirm that it is no longer a problem @ master... @GaelVaroquaux can be closed...

setting multi_class='multinomial' it fails also on master for me. Check the following example

from sklearn.linear_model import LogisticRegressionCV
import numpy as np

n_samples, n_features = 50, 5
rng = np.random.RandomState(0)
X_ref = rng.randn(n_samples, n_features)
y = rng.choice(['foo', 'bar', 'baz'], n_samples)
X_ref -= X_ref.mean()
X_ref /= X_ref.std()
lr_cv = LogisticRegressionCV(Cs=[1.], fit_intercept=False, multi_class='multinomial')
lr_cv.fit(X_ref, y)

Ermm... sorry for suggesting to close this issue... the label encoder was not properly used... fix at #5874