added function thatcalculates average of pr curve for each class

prec_recall.py

@@ -8,7 +8,7 @@ def pr(num_classes, y_test, y_pred):
     recall = dict()
     average_precision = dict()
     for i in range(num_classes):
-        precision[i], recall[i], _ = precision_recall_curve(y_test_one_hot[:, i], y_pred[:, i])
+        precision[i], recall[i], _ = precision_recall_curve(y_test[:, i], y_pred[:, i])
         average_precision[i] = average_precision_score(y_test[:, i], y_pred[:, i])
 
     # A "micro-average": quantifying score on all classes jointly
@@ -32,3 +32,37 @@ def plot_pr(recall, precision, average_precision):
     plt.savefig("precision_recall_curve")
     #plt.show()
     plt.close()
+
+
+def avg_pr(n_splits, num_classes, recs_k, precs_k, avgs_k):
+    prec_per_class = [[precs_k[k][i] for k in range(n_splits)] for i in range(num_classes)]
+    rec_per_class = [[recs_k[k][i] for k in range(n_splits)] for i in range(num_classes)]
+    avg_prec_per_class = [[avgs_k[k][i] for k in range(n_splits)] for i in range(num_classes)]
+
+    # First aggregate all points for every recall curve of one class
+    all_recall_per_class = []
+    for i in range(num_classes):
+        all_recall_per_class.append(np.unique(np.concatenate([recs_k[k][i] for k in range(n_splits)])))
+
+    mean_prec_per_class = []
+    for i in range(num_classes):
+        mean_prec_per_class.append(np.zeros_like(all_recall_per_class[i]))
+
+    avg_prec = np.zeros(num_classes)
+    for i in range(num_classes): # for a determinated class
+        for k in range(n_splits):
+            mean_prec_per_class[i] += np.interp(all_recall_per_class[i], rec_per_class[i][k][::-1],prec_per_class[i][k][::-1])
+            avg_prec[i]+= avg_prec_per_class[i][k]
+            print(avg_prec[i])
+        mean_prec_per_class[i] /= n_splits
+        avg_prec[i] /= n_splits
+
+        plt.figure()
+        plt.step(all_recall_per_class[i], mean_prec_per_class[i], color='b', alpha=0.2, where='post')
+        plt.fill_between(all_recall_per_class[i], mean_prec_per_class[i], alpha=0.2, color='b')
+        plt.xlabel('Recall')
+        plt.ylabel('Precision')
+        plt.ylim([0.0, 1.05])
+        plt.xlim([0.0, 1.0])
+        plt.title('Average precision score, over class {0}: AP={1:0.2f}'.format(i, avg_prec[i]))
+        plt.savefig("pr_curve_class_" +str(i))