Clever Geek Handbook

Generating ROC curve using python for multiclassification

Following from here: Convert 1D array to matrix based on 2D class in python

I want to draw ROC curves for each of my 46 classes. I have 300 test cases that I run my classifier for to make a prediction.

y_test

- these are true classes, and y_pred

- this is what my classifier predicted.

Here's my code:

    from sklearn.metrics import confusion_matrix, roc_curve, auc
    from sklearn.preprocessing import label_binarize
    import numpy as np

    y_test_bi = label_binarize(y_test, classes=[0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18, 19,20,21,2,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,3,40,41,42,43,44,45])
    y_pred_bi = label_binarize(y_pred, classes=[0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18, 19,20,21,2,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,3,40,41,42,43,44,45])
    # Compute ROC curve and ROC area for each class
    fpr = dict()
    tpr = dict()
    roc_auc = dict()
    for i in range(2):
        fpr[i], tpr[i], _ = roc_curve(y_test_bi, y_pred_bi)
        roc_auc[i] = auc(fpr[i], tpr[i])

However, I am now getting the following error:

Traceback (most recent call last):
  File "C:\Users\app\Documents\Python Scripts\gbc_classifier_test.py", line 152, in <module>
    fpr[i], tpr[i], _ = roc_curve(y_test_bi, y_pred_bi)
  File "C:\Users\app\Anaconda\lib\site-packages\sklearn\metrics\metrics.py", line 672, in roc_curve
    fps, tps, thresholds = _binary_clf_curve(y_true, y_score, pos_label)
  File "C:\Users\app\Anaconda\lib\site-packages\sklearn\metrics\metrics.py", line 505, in _binary_clf_curve
    y_true = column_or_1d(y_true)
  File "C:\Users\app\Anaconda\lib\site-packages\sklearn\utils\validation.py", line 265, in column_or_1d
    raise ValueError("bad input shape {0}".format(shape))
ValueError: bad input shape (300L, 46L)
+3


source to share


1 answer


roc_curve

takes a parameter with a form [n_samples]

( link ), and your inputs ( y_test_bi

either y_pred_bi

) are forms (300, 46)

. Pay attention to the first

I think the problem y_pred_bi

is the array of probabilities created by the call clf.predict_proba(X)

(please confirm this). Since your classifier has been trained in all 46 grades, it outputs 46-dimensional vectors for each data point and cannot do anything label_binarize

.



I know two ways to get around this:

  • Place 46 binary classifiers by calling label_binarize

    before clf.fit()

    and then compute the ROC curve
  • Dump each column of the output array 300 by 46 and pass that as the second parameter roc_curve

    . This is my preferred approach, I assume it y_pred_bi

    contains probabilities
+3


source






More articles:


All Articles