Multilabel-indicator is not supported for confusion matrix

multilabel-indicator is not supported is the error message I get, when trying to run:

confusion_matrix(y_test, predictions)

y_test is a DataFrame which is of shape:

Horse | Dog | Cat
1 0 0
0 1 0
0 1 0
... ... ...

predictions is a numpy array:

[[1, 0, 0], [0, 1, 0], [0, 1, 0]]

I've searched a bit for the error message, but haven't really found something I could apply. Any hints?

4 Answers

No, your input to confusion_matrix must be a list of predictions, not OHEs (one hot encodings). Call argmax on your y_test and y_pred, and you should get what you expect.

confusion_matrix( y_test.values.argmax(axis=1), predictions.argmax(axis=1))
array([[1, 0], [0, 2]])

The confusion matrix takes a vector of labels (not the one-hot encoding). You should run

confusion_matrix(y_test.values.argmax(axis=1), predictions.argmax(axis=1))

from sklearn.metrics import confusion_matrix
predictions_one_hot = model.predict(test_data)
cm = confusion_matrix(labels_one_hot.argmax(axis=1), predictions_one_hot.argmax(axis=1))
print(cm)

Output would be something like this:

[[298 2 47 15 77 3 49] [ 14 31 2 0 5 1 2] [ 64 5 262 22 94 38 43] [ 16 1 20 779 15 14 34] [ 49 0 71 33 316 7 118] [ 14 0 42 23 5 323 9] [ 20 1 27 32 97 13 436]]

If you have numpy.ndarray you can try the following

import seaborn as sns
T5_lables = ['4TCM','WCM','WSCCM','IWCM','CCM']
ax= plt.subplot()
cm = confusion_matrix(np.asarray(Y_Test).argmax(axis=1), np.asarray(Y_Pred).argmax(axis=1))
sns.heatmap(cm, annot=True, fmt='g', ax=ax); #annot=True to annotate cells, ftm='g' to disable scientific notation
# labels, title and ticks
ax.set_xlabel('Predicted labels');ax.set_ylabel('True labels');
ax.set_title('Confusion Matrix');
ax.xaxis.set_ticklabels(T5_lables); ax.yaxis.set_ticklabels(T5_lables);