From 48a781cd6aca7be40821b55679bebb30295d4336 Mon Sep 17 00:00:00 2001
From: lixinran <lixr423@outlook.com>
Date: Thu, 31 Dec 2020 10:02:23 +0800
Subject: [PATCH] Revised according to comments

---
 mmcls/core/evaluation/mean_ap.py | 19 ++++++++++++++-----
 1 file changed, 14 insertions(+), 5 deletions(-)

diff --git a/mmcls/core/evaluation/mean_ap.py b/mmcls/core/evaluation/mean_ap.py
index 41fd649a0..c482e1b1b 100644
--- a/mmcls/core/evaluation/mean_ap.py
+++ b/mmcls/core/evaluation/mean_ap.py
@@ -5,6 +5,15 @@ import torch
 def average_precision(pred, target):
     """ Calculate the average precision for a single class
 
+    AP summarizes a precision-recall curve as the weighted mean of maximum
+    precisions obtained for any r'>r, where r is the recall:
+
+    ..math::
+        \\text{AP} = \\sum_n (R_n - R_{n-1}) P_n
+
+    Note that no approximation is involved since the curve is piecewise
+    constant.
+
     Args:
         pred (np.ndarray): The model prediction with shape (N, ).
         target (np.ndarray): The target of each prediction with shape (N, ).
@@ -19,17 +28,17 @@ def average_precision(pred, target):
     sort_target = target[sort_inds]
 
     # count true positive examples
-    p_inds = sort_target == 1
-    tp = np.cumsum(p_inds)
-    total_p = tp[-1]
+    pos_inds = sort_target == 1
+    tp = np.cumsum(pos_inds)
+    total_pos = tp[-1]
 
     # count not difficult examples
     pn_inds = sort_target != -1
     pn = np.cumsum(pn_inds)
 
-    tp[np.logical_not(p_inds)] = 0
+    tp[np.logical_not(pos_inds)] = 0
     precision = tp / np.maximum(pn, eps)
-    ap = np.sum(precision) / np.maximum(total_p, eps)
+    ap = np.sum(precision) / np.maximum(total_pos, eps)
     return ap