Update qfedavg.py

fedlab/contrib/algorithm/qfedavg.py

@@ -71,3 +71,41 @@ def train(self, model_parameters, train_loader) -> None:
         self.hk = self.q * np.float_power(
             ret_loss + 1e-10, self.q - 1) * grad.norm(
             )**2 + 1.0 / self.lr * np.float_power(ret_loss + 1e-10, self.q)
+
+class qFedAvgSerialClientTrainer(SGDSerialClientTrainer):
+    def setup_optim(self, epochs, batch_size, lr, q):
+        super().setup_optim(epochs, batch_size, lr)
+        self.q = q
+
+    def train(self, model_parameters, train_loader) -> None:
+        """Client trains its local model on local dataset.
+        Args:
+            model_parameters (torch.Tensor): Serialized model parameters.
+        """
+        self.set_model(model_parameters)
+        # self._LOGGER.info("Local train procedure is running")
+        for ep in range(self.epochs):
+            self._model.train()
+            ret_loss = 0.0
+            for data, target in train_loader:
+                if self.cuda:
+                    data, target = data.cuda(self.device), target.cuda(
+                        self.device)
+
+                outputs = self._model(data)
+                loss = self.criterion(outputs, target)
+
+                self.optimizer.zero_grad()
+                loss.backward()
+                self.optimizer.step()
+
+                ret_loss += loss.detach().item()
+        # self._LOGGER.info("Local train procedure is finished")
+
+        grad = (model_parameters - self.model_parameters) / self.lr
+        self.delta = grad * np.float_power(ret_loss + 1e-10, self.q)
+        self.hk = self.q * np.float_power(
+            ret_loss + 1e-10, self.q - 1) * grad.norm(
+            )**2 + 1.0 / self.lr * np.float_power(ret_loss + 1e-10, self.q)
+
+        return [self.delta, self.hk]