-2

I wanted to plot ROC curve for my classification model. As I am new to this, I read about it, saw few posts and referring this SO answer I created a roc curve.

My data is this type:

print(Y.shape)
print(predictions.shape)
print(Y)
print(predictions)

(1, 400)
(1, 400)
[[0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
  0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
  0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
  0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
  0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
  0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
  1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
  1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
  1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
  1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
  1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
  1 1 1 1]]
[[0 1 1 1 0 1 0 1 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
  0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0
  0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
  0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
  0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
  0 0 0 0 0 0 0 0 0 0 1 0 1 1 1 0 1 1 1 1 0 1 1 1 0 1 0 0 1 1 1 1 1 1 0 0
  1 1 1 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
  1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 1 1 1 1 1 1 1 1 1 0 1
  0 1 1 1 1 1 1 1 1 1 1 1 1 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
  1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 1 0 1 1 1
  1 1 1 1 1 0 1 1 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 1 1 1 1 1 1 0 1 0
  0 0 1 0]]

After implementing the code:

from sklearn.metrics import precision_score
print('Precsion score: '+ str(precision_score(Y.ravel(), predictions.ravel())))

from sklearn.metrics import recall_score
print('Recall score: '+ str(recall_score(Y.ravel(), predictions.ravel())))

from sklearn.metrics import f1_score
print('F1 score: '+ str(f1_score(Y.ravel(), predictions.ravel())))

from sklearn.metrics import roc_auc_score, auc, roc_curve
print('ROC score: ' + str(roc_auc_score(Y.ravel(), predictions.ravel())))

from sklearn.metrics import confusion_matrix
print('Confusion matrix: ')
print(confusion_matrix(Y.ravel(), predictions.ravel()))

fpr = dict()
tpr = dict()
threshold = dict()
roc_auc = dict()

for i in range(2):
    fpr[i], tpr[i], threshold[i] = roc_curve(Y.ravel(), predictions.ravel())
    roc_auc[i] = auc(fpr[i], tpr[i])
print(fpr, tpr, threshold, roc_auc)

plt.figure()
plt.plot(fpr[1], tpr[1])
plt.xlabel('False Positive Rate')
plt.ylabel('True Positive Rate')
plt.title('Receiver operating characteristic(ROC Curve)')

Output:

Precsion score: 0.9179487179487179
Recall score: 0.895
F1 score: 0.9063291139240507
ROC score: 0.9075
Confusion matrix: 
[[184  16]
 [ 21 179]]
{0: array([0.  , 0.08, 1.  ]), 1: array([0.  , 0.08, 1.  ])} {0: array([0.   , 0.895, 1.   ]), 1: array([0.   , 0.895, 1.   ])} {0: array([2, 1, 0]), 1: array([2, 1, 0])} {0: 0.9075, 1: 0.9075}
Text(0.5, 1.0, 'Receiver operating characteristic(ROC Curve)')

ROC Curve

I didn't understand why loop was used? I can see three values calculate in each row for FPR, TPR, Threshold and roc_auc. I did read that roc_curve takes probabilities as target scores(I will work on that). However I am not able to follow from my inputting (1,400)dimension data to how these array got calculated?

Thanks in advance.

desertnaut
  • 57,590
  • 26
  • 140
  • 166
Deepak Dhiman
  • 51
  • 7
  • 2
    As the linked SO answers says clearly, you need probabilistic predictions in order to plot a ROC curve, *not* hard classes 0/1 as you do here. – desertnaut Sep 03 '20 at 09:36

1 Answers1

1

I do not understand either, why you used the loop, because you can have the same functionality of the code by removing it and adjusting your code:

import matplotlib.pyplot as plt
from sklearn.metrics import precision_score
from sklearn.metrics import recall_score
from sklearn.metrics import f1_score
from sklearn.metrics import roc_auc_score, auc, roc_curve
from sklearn.metrics import confusion_matrix

Y=[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
       1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
       1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
       1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
       1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
       1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
       1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
       1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
       1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
       1, 1, 1, 1]

predictions=[0, 1, 1, 1, 0, 1, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 1, 1, 0, 1, 1,
       1, 1, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1, 1, 0,
       1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
       1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
       1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1,
       0, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1,
       1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
       1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
       1, 1, 0, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 0, 1, 1, 1, 1, 1,
       1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 0, 1, 0,
       0, 0, 1, 0]

print('Precsion score: '+ str(precision_score(Y, predictions)))
print('Recall score: '+ str(recall_score(Y, predictions)))
print('F1 score: '+ str(f1_score(Y, predictions)))
print('ROC score: ' + str(roc_auc_score(Y, predictions)))
print('Confusion matrix: ')
print(confusion_matrix(Y, predictions))

fpr, tpr, threshold = roc_curve(Y, predictions)
roc_auc = auc(fpr, tpr)

print(fpr, tpr, threshold, roc_auc)

plt.figure()
plt.plot(fpr, tpr)
plt.xlabel('False Positive Rate')
plt.ylabel('True Positive Rate')
plt.title('Receiver operating characteristic(ROC Curve)')

It yields the output:

Precsion score: 0.9179487179487179
Recall score: 0.895
F1 score: 0.9063291139240507
ROC score: 0.9075
Confusion matrix: 
[[184  16]
 [ 21 179]]
[0.   0.08 1.  ] [0.    0.895 1.   ] [2 1 0] 0.9075

enter image description here

Your 400 data points are used to compute the ROC curve, but only three data points are present in the visualization, because you only have two unique values in your data (0 and 1).

To quote the answer from here:

The number of points depend on the number of unique values in the input. Since the input vector has only 2 unique values, the function gives correct output.

Kim Tang
  • 2,330
  • 2
  • 9
  • 34
  • 2
    A reminder that ROC curve requires probabilistic predictions and not hard classes 0/1 should arguably be in place in the answer. – desertnaut Sep 03 '20 at 09:37
  • @desertnaut To my understanding ROC works with any model prediction output as long as it can be ranked, not only probabilistic predictions. If the output of the model is continuous, which you then artificially make discrete (0, 1), and you then rank based on (0, 1), you can not accurately rank your scores anymore. That's the reason why you should not calculate ROC with 0/1 predicted labels. – Joop Jan 19 '22 at 13:08
  • See also: https://scikit-learn.org/stable/modules/model_evaluation.html#roc-metrics – Joop Jan 19 '22 at 13:11