ImportError: No module named model_selection

Question

I am trying to use train_test_split function and write:

from sklearn.model_selection import train_test_split

and this causes

ImportError: No module named model_selection

Why? And how to overcome?

Answer 1

I guess you have the wrong version of scikit-learn, a similar situation was described here on GitHub. Previously (before v0.18), train_test_split was located in the cross_validation module:

from sklearn.cross_validation import train_test_split

However, now it's in the model_selection module:

from sklearn.model_selection import train_test_split

so you'll need the newest version.

To upgrade to at least version 0.18, do:

pip install -U scikit-learn

(Or pip3, depending on your version of Python). If you've installed it in a different way, make sure you use another method to update, for example when using Anaconda.

Answer 2

Update sklearn:

conda update scikit-learn

Answer 3

I encountered this problem when I import GridSearchCV.

Just changed sklearn.model_selection to sklearn.grid_search.

Answer 4

I had the same problem while using Jupyter Notebook, no matter what I updated in Python 3, conda, I could not get in Jupyter:

import sklearn
print (sklearn.__version__)
0.17.1

to SHOW scikit-learn-0.18.1

Finally, I removed Anaconda3 and Jupyter Notebook and reinstalled fresh. I got it to work.

http://ukitech.blogspot.com/2017/02/sklearnmodelselection.html

Answer 5

Do you have sklearn?

If not, do the following:

sudo pip install sklearn

After installing sklearn:

from sklearn.model_selection import train_test_split

works fine.

Answer 6

To install scikit-learn version 18.0, I used both commands:

conda update scikit-learn

pip install -U scikit-learn

But it does not work. There was a problem "Cannot install 'scikit-learn'. It is a distutils installed project and thus we cannot accurately determine which files belong to it which would lead to only a partial uninstall".

Finally, i can install it by using following command:

pip install --user --upgrade scikit-learn==0.18.0

Answer 7

In Late September 2016, SciKit Learn 0.18 was released and there was a slight change to the code. With SciKit Learn 0.18 the train_test_split function is now imported from model_selection instead of cross_validation.

from sklearn.cross_validation import train_test_split

has been changed to :

from sklearn.model_selection import train_test_split

The same has also happened for GridSearchCV.

Answer 8

As @linusg said, one option is just import crossvalidation as follows:

from sklearn import cross_validation
X_train,X_test,y_train,y_test = cross_validation.train_test_split(X,y,test_size=0.3)

Answer 9

Latest Stable release of sklearn 0.20.0 has train_test_split under model_selection not under cross_validation.

In order to check your sklearn version:

import sklearn
print (sklearn.__version__)
0.20.2

Answer 10

Adding some info to the previous answer from @linusg :

sklearn keeps a release history of all its changes. Think of checking it from time to time. Here is the link to the documentation.

As you can see in the documentation for the version 0.18, a new module was created called model_selection. Therefore it didn't exist in previous versions.

Update sklearn and it will work !

Answer 11

Your sklearn version is too low, model_selection is imported by 0.18.1, so please update the sklearn version.