add dy-lr in lenet alexnet

example/alexnet_cifar10/generator_lr.py

@@ -0,0 +1,44 @@
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+"""learning rate generator"""
+import numpy as np
+
+
+def get_lr(current_step, lr_max, total_epochs, steps_per_epoch):
+    """
+    generate learning rate array
+
+    Args:
+       current_step(int): current steps of the training
+       lr_max(float): max learning rate
+       total_epochs(int): total epoch of training
+       steps_per_epoch(int): steps of one epoch
+
+    Returns:
+       np.array, learning rate array
+    """
+    lr_each_step = []
+    total_steps = steps_per_epoch * total_epochs
+    decay_epoch_index = [0.8 * total_steps]
+    for i in range(total_steps):
+        if i < decay_epoch_index[0]:
+            lr = lr_max
+        else:
+            lr = lr_max * 0.1
+        lr_each_step.append(lr)
+    lr_each_step = np.array(lr_each_step).astype(np.float32)
+    learning_rate = lr_each_step[current_step:]
+
+    return learning_rate

example/alexnet_cifar10/train.py

@@ -21,12 +21,14 @@ python train.py --data_path /YourDataPath
 import argparse
 from config import alexnet_cfg as cfg
 from dataset import create_dataset
+from generator_lr import get_lr
 import mindspore.nn as nn
 from mindspore import context
+from mindspore import Tensor
 from mindspore.train import Model
 from mindspore.nn.metrics import Accuracy
 from mindspore.model_zoo.alexnet import AlexNet
-from mindspore.train.callback import ModelCheckpoint, CheckpointConfig, LossMonitor
+from mindspore.train.callback import ModelCheckpoint, CheckpointConfig, LossMonitor, TimeMonitor
 
 
 if __name__ == "__main__":
@@ -43,16 +45,17 @@ if __name__ == "__main__":
 
     network = AlexNet(cfg.num_classes)
     loss = nn.SoftmaxCrossEntropyWithLogits(is_grad=False, sparse=True, reduction="mean")
-    opt = nn.Momentum(network.trainable_params(), cfg.learning_rate, cfg.momentum)
+    lr = Tensor(get_lr(0, cfg.learning_rate, cfg.epoch_size, cfg.save_checkpoint_steps))
+    opt = nn.Momentum(network.trainable_params(), lr, cfg.momentum)
     model = Model(network, loss, opt, metrics={"Accuracy": Accuracy()})  # test
 
     print("============== Starting Training ==============")
     ds_train = create_dataset(args.data_path,
                               cfg.batch_size,
-                              cfg.epoch_size,
-                              "train")
+                              cfg.epoch_size)
+    time_cb = TimeMonitor(data_size=ds_train.get_dataset_size())
     config_ck = CheckpointConfig(save_checkpoint_steps=cfg.save_checkpoint_steps,
                                  keep_checkpoint_max=cfg.keep_checkpoint_max)
     ckpoint_cb = ModelCheckpoint(prefix="checkpoint_alexnet", directory=args.ckpt_path, config=config_ck)
-    model.train(cfg.epoch_size, ds_train, callbacks=[ckpoint_cb, LossMonitor()],
+    model.train(cfg.epoch_size, ds_train, callbacks=[time_cb, ckpoint_cb, LossMonitor()],
                 dataset_sink_mode=args.dataset_sink_mode)

example/lenet_mnist/train.py

@@ -25,7 +25,7 @@ from dataset import create_dataset
 import mindspore.nn as nn
 from mindspore.model_zoo.lenet import LeNet5
 from mindspore import context
-from mindspore.train.callback import ModelCheckpoint, CheckpointConfig, LossMonitor
+from mindspore.train.callback import ModelCheckpoint, CheckpointConfig, LossMonitor, TimeMonitor
 from mindspore.train import Model
 from mindspore.nn.metrics import Accuracy
 
@@ -40,19 +40,20 @@ if __name__ == "__main__":
 
     args = parser.parse_args()
 
-    context.set_context(mode=context.GRAPH_MODE, device_target=args.device_target)
+    context.set_context(mode=context.GRAPH_MODE, device_target=args.device_target, enable_mem_reuse=False)
+    ds_train = create_dataset(os.path.join(args.data_path, "train"),
+                              cfg.batch_size,
+                              cfg.epoch_size)
 
     network = LeNet5(cfg.num_classes)
     net_loss = nn.SoftmaxCrossEntropyWithLogits(is_grad=False, sparse=True, reduction="mean")
     net_opt = nn.Momentum(network.trainable_params(), cfg.lr, cfg.momentum)
+    time_cb = TimeMonitor(data_size=ds_train.get_dataset_size())
     config_ck = CheckpointConfig(save_checkpoint_steps=cfg.save_checkpoint_steps,
                                  keep_checkpoint_max=cfg.keep_checkpoint_max)
     ckpoint_cb = ModelCheckpoint(prefix="checkpoint_lenet", config=config_ck)
     model = Model(network, net_loss, net_opt, metrics={"Accuracy": Accuracy()})
 
-    ds_train = create_dataset(os.path.join(args.data_path, "train"),
-                              cfg.batch_size,
-                              cfg.epoch_size)
     print("============== Starting Training ==============")
-    model.train(cfg['epoch_size'], ds_train, callbacks=[ckpoint_cb, LossMonitor()],
+    model.train(cfg['epoch_size'], ds_train, callbacks=[time_cb, ckpoint_cb, LossMonitor()],
                 dataset_sink_mode=args.dataset_sink_mode)