"""roc.py
Provides functions for computing ROC AUC and plotting then saving ROC curves
in a multiclass setting.
Requires: NumPy, SciPy, matplotlib, scikit-learn (and their dependencies)
Author: Ji-Sung Kim, Rzhetsky Lab
Copyright: 2018, all rights reserved
"""
from __future__ import print_function
import numpy as np
import matplotlib
matplotlib.use('Agg') # do not run X server
import matplotlib.pyplot as plt
plt.style.use('ggplot')
from scipy import interp
from sklearn.metrics import roc_curve, auc
from sklearn.preprocessing import label_binarize
def compute_and_plot_roc_scores(y_test, y_test_probas, num_class, path=None):
"""Compute ROC statistics and plot ROC curves.
Arguments:
y_test: [int]
list of test class labels as integer indices
y_test_probas: np.ndarray, float
array of predicted probabilities with shape
(num_sample, num_class)
num_class: int
number of classes
path: string
filepath where to save the ROC curve plot; if None will not perform
plotting
Returns:
roc_auc_dict: {int: float}
dictionary mapping classes to ROC AUC scores
fpr_dict: {string: np.ndarray}
dictionary mapping names of classes or an averaging method to
arrays of increasing false positive rates
tpr_dict: {string: float}
dictionary mapping names of classes or an averaging method to
arrays of increasing true positive rates
"""
roc_auc_dict, fpr_dict, tpr_dict = _compute_roc_stats(y_test, y_test_probas,
num_class)
if path is not None:
_create_roc_plot(roc_auc_dict, fpr_dict, tpr_dict, num_class, path)
return roc_auc_dict, fpr_dict, tpr_dict
def _create_roc_plot(roc_auc_dict, fpr_dict, tpr_dict, num_class, path):
"""Create and save a combined ROC plot to file.
Arguments:
roc_auc_dict: {int: float}
dictionary mapping classes to ROC AUC scores
fpr_dict: {string: np.ndarray}
dictionary mapping names of classes or an averaging method to
arrays of increasing false positive rates
tpr_dict: {string: float}
dictionary mapping names of classes or an averaging method to
arrays of increasing true positive rates
num_class: int
number of classes
path: string
filepath where to save the plot
"""
# aggregate all false positive rates
all_fpr = np.unique(np.concatenate(
[fpr_dict[i] for i in range(num_class)]))
# interpolate all ROC curves at this points
mean_tpr = np.zeros_like(all_fpr)
for i in range(num_class):
mean_tpr += interp(all_fpr, fpr_dict[i], tpr_dict[i])
# average and compute AUC
mean_tpr /= num_class
fpr_dict["macro"] = all_fpr
tpr_dict["macro"] = mean_tpr
roc_auc_dict["macro"] = auc(fpr_dict["macro"], tpr_dict["macro"])
# plot
plt.figure()
plt.plot(fpr_dict["micro"], tpr_dict["micro"],
label='micro-average ROC curve (area = {0:0.2f})'.format(
roc_auc_dict["micro"]),
linewidth=2)
plt.plot(fpr_dict["macro"], tpr_dict["macro"],
label='macro-average ROC curve (area = {0:0.2f})'.format(
roc_auc_dict["macro"]),
linewidth=2)
for i in range(num_class):
plt.plot(fpr_dict[i], tpr_dict[i],
label='ROC curve of class {0} (area = {1:0.2f})'.format(
i, roc_auc_dict[i]))
plt.plot([0, 1], [0, 1], 'k--')
plt.xlim([0.0, 1.0])
plt.ylim([0.0, 1.05])
plt.xlabel('False Positive Rate')
plt.ylabel('True Positive Rate')
plt.title('Some extension of Receiver operating characteristic to multi-class')
plt.legend(loc="lower right")
plt.savefig(path) # save plot
plt.close()
def _compute_roc_stats(y_test, y_test_probas, num_class):
"""Compute ROC AUC statistics and visualize ROC curves.
Arguments:
y_test: [int]
list of test class labels as integer indices
y_test_probas: np.ndarray, float
array of predicted probabilities with shape
(num_sample, num_class)
num_class: int
number of classes
Returns:
roc_auc_dict: {int: float}
dictionary mapping classes to ROC AUC scores
fpr_dict: {string: np.ndarray}
dictionary mapping names of classes or an averaging method to
arrays of increasing false positive rates
tpr_dict: {string: float}
dictionary mapping names of classes or an averaging method to
arrays of increasing true positive rates
"""
y_test = label_binarize(y_test, classes=range(0, num_class))
fpr_dict, tpr_dict, roc_auc_dict = {}, {}, {}
for i in range(num_class):
fpr_dict[i], tpr_dict[i], _ = roc_curve(
y_test[:, i], y_test_probas[:, i])
roc_auc_dict[i] = auc(fpr_dict[i], tpr_dict[i])
# Compute micro-average ROC curve and ROC area
fpr_dict["micro"], tpr_dict["micro"], _ = roc_curve(
y_test.ravel(), y_test_probas.ravel())
roc_auc_dict["micro"] = auc(fpr_dict["micro"], tpr_dict["micro"])
return roc_auc_dict, fpr_dict, tpr_dict
