Scikit multi-class classification metrics, classification report

Question

I am using scikit learn 0.15.2 for a multi-class classification problem. I was getting a lot of DeprecationWarnings as follows when following examples like: scikit 0.14 multi label metrics until I started to use the MultiLabelBinarizer:

"DeprecationWarning: Direct support for sequence of sequences multilabel representation will be unavailable from version 0.17. Use sklearn.preprocessing.MultiLabelBinarizer to convert to a label indicator representation."

However, I cannot find a way to get the classification report (with precision, recall, f-measure) to work with it, as i was previously possible as shown here: scikit 0.14 multi label metrics

I tried to use inverse_transform as below, this gives a classification_report but also gives the warnings again, that from 0.17 this code will break.

How can I get measures for a multi-class classification problem?

Example code:

import numpy as np
from sklearn.multiclass import OneVsRestClassifier
from sklearn.preprocessing import MultiLabelBinarizer
from sklearn.svm import LinearSVC
from sklearn.metrics import classification_report

# Some simple data:

X_train = np.array([[0,0,0], [0,0,1], [0,1,0], [1,0,0], [1,1,1]])
y_train = [[1], [1], [1,2], [2], [2]]

# Use MultiLabelBinarizer and train a multi-class classifier:

mlb = MultiLabelBinarizer(sparse_output=True)
y_train_mlb = mlb.fit_transform(y_train)

clf = OneVsRestClassifier(LinearSVC())
clf.fit(X_train, y_train_mlb)

# classification_report, here I did not find a way to use y_train_mlb, 
# I am getting a lot of DeprecationWarnings

predictions_test = mlb.inverse_transform(clf.predict(X_train))
print classification_report(y_train, predictions_test)

# Predict new example:

print mlb.inverse_transform(clf.predict(np.array([0,1,0])))

Answer 1

It seems like you have to run your classification report with the binarized labels:

print classification_report(y_train_mlb, clf.predict(X_train))