2x2 confusion matrix in three-class classification?

Question

I am doing classification on a dataset with three classes (Labels Low, Medium, High).

I run the following code to get my confusion matrix:

from sklearn.metrics import confusion_matrix
cm = confusion_matrix(y_test, y_pred)

And I get the following output for cm:

array([[18, 10],
       [ 7, 61]], dtype=int64)

What does this output mean? I read the following link but didn't understood Confusion Matrix and Class Statistics

how about [scikit docs](https://scikit-learn.org/stable/modules/model_evaluation.html#confusion-matrix)? — alex, Mar 05 '20 at 10:07
[tag:spyder] has nothing at all to do with this, it's only an IDE, it's not going to affect how your code runs. The relevant tags are [tag:scikit-learn], [tag:classification], [tag:multiclass-classification] — smci, Mar 05 '20 at 10:26
If you have three different classes in `y`, then confusion matrix should have been 3x3. Could you check if all labels exist in `y_test` and `y_pred`? — Ala Tarighati, Mar 05 '20 at 10:30
You say you have three classes, but it looks like your classifier only did two-class. Post us the code you used for training your classifier. In particular make sure you're passing the label column in correctly. — smci, Mar 05 '20 at 10:31
You haven't posted the code for your classifier, so we can't solve why it only trained to two classes not three. SO rules require you to [Minimal, Complete Verifiable Example (MCVE)](https://stackoverflow.com/help/minimal-reproducible-example) — smci, Mar 06 '20 at 01:44

seralouk · Answer 1 · 2020-03-05T10:13:52.790

-1

This is the explanation of what you get.

For any NxN confusion matrix the following holds:

Actual: is the true labels

For a 2x2 case (simpler explanation):

edited Mar 05 '20 at 10:13

answered Mar 05 '20 at 10:08

seralouk

Ala Tarighati · Answer 2 · 2020-03-05T10:21:23.873

Let us take a closer look at the results:

tn, fp, fn, tp = np.array([[18, 10],
                           [ 7, 61]]).ravel()

Which means:

tn (True Negative) = 18
fp (False Positive) = 10
fn (False Negative) = 7 
tp (True Positive) = 61

In other words,

You could also directly use

tn, fp, fn, tp = confusion_matrix(y_true, y_pred).ravel()

2 Answers2