Different results for LogisticRegression on python 2.7 and 3

Assuming that you are using sklearn, if you go to the source in the github you will see that they do not differ.

But you have variability elsewhere. For example train_test_split of sklearn also takes random seed, so it could be that your data is fundamentally different between these approaches (you could do cross valdiation for example to include all the data in train and test)

The main problem was the ordinal data transformation. Python 2.7 somehow has detected the ordinal data and works with it. In python3 I have to convert the ordinal data with:

from sklearn.preprocessing import OrdinalEncoder
encoder = OrdinalEncoder()
train_df.F1 = encoder.fit_transform(train_df.F1.values.reshape(-1, 1))

Now it works with python3 also!