Pickling sklearn models and using them in web app

Question

I have created an sklearn model that I would like to imbed into my nodejs web app. I have created the model with python3 and packaged it into a pkl file with joblib.

from sklearn.externals import joblib
joblib.dump(clf, 'RandomForest_jul30.pkl', protocol=2)

I am using npm package "child_process" to run a python script which I will load the data into and unpackage the pkl file with joblib.loads although it throws the error:

ValueError: unsupported pickle protocol: 3

I have tried to change which python version I am using although I still get similar errors or version related errors.

Edit: when run with python 3 the error is:

/usr/lib/python3/dist-packages/sklearn/base.py:315: UserWarning: Trying to unpickle estimator DecisionTreeClassifier from version 0.18.1 when using version 0.18. This might lead to breaking code or invalid results. Use at your own risk.
  UserWarning) . . . KeyError: 2

score 0 · Answer 1 · answered Aug 07 '17 at 16:15

0

Try using cPickle

import cPickle as pkl
pkl.dump(clf,open('Clf.clf','wb+'))

answered Aug 07 '17 at 16:15

Mohamed Emad

123
1
14

I am a little confused. Some resources are saying cPickle is outdated and or is no longer cPickle – Christopher Evans Aug 07 '17 at 16:46
in python3 it is called _pickle it is just the normal pickle but faster – Mohamed Emad Aug 07 '17 at 19:33
Is unpacking files with this library the same as unpacking with joblib or pickle? – Christopher Evans Aug 08 '17 at 02:31

Pickling sklearn models and using them in web app

1 Answers1