From e001d34c0c6ebd73bd452c7fd2152e45c1a21db5 Mon Sep 17 00:00:00 2001
From: zwl <15878005+zwl90188229@user.noreply.gitee.com>
Date: Tue, 15 Jul 2025 12:49:33 +0000
Subject: [PATCH] add StreamLearn/Algorithm/PAA/paa_proto.py.

Signed-off-by: zwl <15878005+zwl90188229@user.noreply.gitee.com>
---
 StreamLearn/Algorithm/PAA/paa_proto.py | 20 ++++++++++++++++++++
 1 file changed, 20 insertions(+)
 create mode 100644 StreamLearn/Algorithm/PAA/paa_proto.py

diff --git a/StreamLearn/Algorithm/PAA/paa_proto.py b/StreamLearn/Algorithm/PAA/paa_proto.py
new file mode 100644
index 0000000..af9af97
--- /dev/null
+++ b/StreamLearn/Algorithm/PAA/paa_proto.py
@@ -0,0 +1,20 @@
+import torch
+import numpy as np
+
+class OnlinePrototypeLearning:
+    def __init__(self, ema_alpha=0.9):
+        self.ema_alpha = ema_alpha
+        self.prototypes = None  # 存储类原型：shape=[num_classes, feat_dim]
+
+    def update_prototypes(self, feats, labels):
+        """根据当前batch的特征和标签更新原型"""
+        feats = feats.detach().cpu().numpy()
+        labels = labels.detach().cpu().numpy()
+        for c in np.unique(labels):
+            mask = (labels == c)
+            class_feats = feats[mask]
+            if self.prototypes is None:
+                self.prototypes = {c: np.mean(class_feats, axis=0)}
+            else:
+                # EMA更新：原型 = alpha*旧原型 + (1-alpha)*新均值
+                self.prototypes[c] = self.ema_alpha * self.prototypes[c] + (1 - self.ema_alpha) * np.mean(class_feats, axis=0)
\ No newline at end of file
-- 
Gitee