[Docs] Add a document about setting interval (#964)

* Add a document about setting interval. Add a document about Setting the Frequency of Logging, Weight Saving, and Validation. * Update set_interval.md Fixed a small bug in the hyperlink. * Update set_interval.md Fixed the wrong hyperlink * Update with pre-commit. * Update docs/en/common_usage/set_interval.md Co-authored-by: Zaida Zhou <58739961+zhouzaida@users.noreply.github.com> * Update docs/en/common_usage/set_interval.md Co-authored-by: Zaida Zhou <58739961+zhouzaida@users.noreply.github.com> * Update docs/en/common_usage/set_interval.md Co-authored-by: Zaida Zhou <58739961+zhouzaida@users.noreply.github.com> * Update docs/en/common_usage/set_interval.md Co-authored-by: Zaida Zhou <58739961+zhouzaida@users.noreply.github.com> * Update docs/en/common_usage/set_interval.md Co-authored-by: Zaida Zhou <58739961+zhouzaida@users.noreply.github.com> * Update docs/en/common_usage/set_interval.md Co-authored-by: Zaida Zhou <58739961+zhouzaida@users.noreply.github.com> * Update docs/en/common_usage/set_interval.md Co-authored-by: Zaida Zhou <58739961+zhouzaida@users.noreply.github.com> * Update docs/en/common_usage/set_interval.md Co-authored-by: Zaida Zhou <58739961+zhouzaida@users.noreply.github.com> * Update docs/en/common_usage/set_interval.md Co-authored-by: Zaida Zhou <58739961+zhouzaida@users.noreply.github.com> * Update docs/en/common_usage/set_interval.md Co-authored-by: Zaida Zhou <58739961+zhouzaida@users.noreply.github.com> * Update docs/en/common_usage/set_interval.md Co-authored-by: Zaida Zhou <58739961+zhouzaida@users.noreply.github.com> * Update docs/en/common_usage/set_interval.md Co-authored-by: Zaida Zhou <58739961+zhouzaida@users.noreply.github.com> * Update index.rst Added: common_usage/set_interval.md --------- Co-authored-by: YuetianW <wangyuetian@supermicro.dscvlab.com> Co-authored-by: Zaida Zhou <58739961+zhouzaida@users.noreply.github.com>
open-mmlab · Mar 1, 2023 · 53ae4fb · 53ae4fb
1 parent cb7e04d
commit 53ae4fb
Show file tree

Hide file tree

Showing 2 changed files with 117 additions and 0 deletions.
diff --git a/docs/en/common_usage/set_interval.md b/docs/en/common_usage/set_interval.md
@@ -0,0 +1,116 @@
+# Setting the Frequency of Logging, Checkpoint Saving, and Validation
+
+MMEngine supports two training modes, `EpochBased` based on epochs and `IterBased` based on the number of iterations. Both of these modes are used in downstream algorithm libraries such as [MMDetection](https://github.com/open-mmlab/mmdetection), which uses the `EpochBased` mode by default, and [MMSegmentation](https://github.com/open-mmlab/mmsegmentation), which uses the `IterBased` mode by default.
+
+In different training modes, the semantics of the interval in MMEngine will be different. In EpochBased mode, the interval is in terms of epochs, while in IterBased mode, the interval is in terms of iterations.
+
+## Setting the Interval for Training and Validation
+
+To customize the interval for training and validation, set the `val_interval` parameter in the initialization parameter `train_cfg` of [Runner](mmengine.runner.Runner).
+
+- EpochBased
+
+In `EpochBased` mode, the default value of `val_interval` is 1, which means to validate once after training an epoch.
+
+```python
+runner = Runner(
+    model=MMResNet50(),
+    work_dir='./work_dir',
+    train_dataloader=train_dataloader,
+    optim_wrapper=dict(optimizer=dict(type=SGD, lr=0.001, momentum=0.9)),
+    train_cfg=dict(by_epoch=True, max_epochs=5, val_interval=1),
+    val_dataloader=val_dataloader,
+    val_cfg=dict(),
+    val_evaluator=dict(type=Accuracy),
+)
+runner.train()
+```
+
+- IterBased
+
+In `IterBased` mode, the default value of `val_interval` is 1000, which means to validate once after training 1000 iterations.
+
+```python
+runner = Runner(
+    model=MMResNet50(),
+    work_dir='./work_dir',
+    train_dataloader=train_dataloader,
+    optim_wrapper=dict(optimizer=dict(type=SGD, lr=0.001, momentum=0.9)),
+    train_cfg=dict(by_epoch=False, max_iters=10000, val_interval=2000),
+    val_dataloader=val_dataloader,
+    val_cfg=dict(),
+    val_evaluator=dict(type=Accuracy),
+)
+runner.train()
+```
+
+## Setting the Interval for Saving Checkpoints
+
+To customize the interval for saving checkpoints, set the `interval` parameter of [CheckpointHook](mmengine.hooks.CheckpointHook).
+
+- EpochBased
+
+In `EpochBased` mode, the default value of `interval` is 1, which means to save checkpoints once after training for one epoch.
+
+```python
+# set the interval to 2, which means to save checkpoints every 2 epochs
+default_hooks = dict(checkpoint=dict(type='CheckpointHook', interval=2))
+runner = Runner(
+    model=MMResNet50(),
+    work_dir='./work_dir',
+    train_dataloader=train_dataloader,
+    optim_wrapper=dict(optimizer=dict(type=SGD, lr=0.001, momentum=0.9)),
+    train_cfg=dict(by_epoch=True, max_epochs=5, val_interval=1),
+    val_dataloader=val_dataloader,
+    val_cfg=dict(),
+    val_evaluator=dict(type=Accuracy),
+    default_hooks=default_hooks,
+)
+runner.train()
+```
+
+- IterBased
+
+By default, checkpoints are saved in terms of epochs. If you want to save checkpoints in terms of iterations, you need to set `by_epoch=False`.
+
+```python
+# set by_epoch=False and interval=500, which means to save checkpoints every 500 iterations
+default_hooks = dict(checkpoint=dict(type='CheckpointHook', by_epoch=False, interval=500))
+runner = Runner(
+    model=MMResNet50(),
+    work_dir='./work_dir',
+    train_dataloader=train_dataloader,
+    optim_wrapper=dict(optimizer=dict(type=SGD, lr=0.001, momentum=0.9)),
+    train_cfg=dict(by_epoch=False, max_iters=10000, val_interval=1000),
+    val_dataloader=val_dataloader,
+    val_cfg=dict(),
+    val_evaluator=dict(type=Accuracy),
+    default_hooks=default_hooks,
+)
+runner.train()
+```
+
+For more information on how to use `CheckpointHook`, please refer to the [CheckpointHook tutorial](../tutorials/hook.md#checkpointhook).
+
+## Setting the Interval for Printing Logs
+
+By default, logs are printed to the terminal once every 10 iterations. You can set the interval using the `interval` parameter of the [LoggerHook](mmengine.hooks.LoggerHook).
+
+```python
+# print logs every 20 iterations
+default_hooks = dict(logger=dict(type='LoggerHook', interval=20))
+runner = Runner(
+    model=MMResNet50(),
+    work_dir='./work_dir',
+    train_dataloader=train_dataloader,
+    optim_wrapper=dict(optimizer=dict(type=SGD, lr=0.001, momentum=0.9)),
+    train_cfg=dict(by_epoch=True, max_epochs=5, val_interval=1),
+    val_dataloader=val_dataloader,
+    val_cfg=dict(),
+    val_evaluator=dict(type=Accuracy),
+    default_hooks=default_hooks,
+)
+runner.train()
+```
+
+For more information on how to use `LoggerHook`, please refer to the [LoggerHook tutorial](../tutorials/hook.md#loggerhook).
diff --git a/docs/en/index.rst b/docs/en/index.rst
@@ -24,6 +24,7 @@ You can switch between Chinese and English documents in the lower-left corner of
    common_usage/speed_up_training.md
    common_usage/save_gpu_memory.md
    common_usage/debug_tricks.md
+   common_usage/set_interval.md
    common_usage/epoch_to_iter.md
 
 .. toctree::