b900fdf6f2
Closes #947 This removes all kwargs from all policy constructors. While doing that, I also improved several names and added a whole lot of TODOs. ## Functional changes: 1. Added possibility to pass None as `critic2` and `critic2_optim`. In fact, the default behavior then should cover the absolute majority of cases 2. Added a function called `clone_optimizer` as a temporary measure to support passing `critic2_optim=None` ## Breaking changes: 1. `action_space` is no longer optional. In fact, it already was non-optional, as there was a ValueError in BasePolicy.init. So now several examples were fixed to reflect that 2. `reward_normalization` removed from DDPG and children. It was never allowed to pass it as `True` there, an error would have been raised in `compute_n_step_reward`. Now I removed it from the interface 3. renamed `critic1` and similar to `critic`, in order to have uniform interfaces. Note that the `critic` in DDPG was optional for the sole reason that child classes used `critic1`. I removed this optionality (DDPG can't do anything with `critic=None`) 4. Several renamings of fields (mostly private to public, so backwards compatible) ## Additional changes: 1. Removed type and default declaration from docstring. This kind of duplication is really not necessary 2. Policy constructors are now only called using named arguments, not a fragile mixture of positional and named as before 5. Minor beautifications in typing and code 6. Generally shortened docstrings and made them uniform across all policies (hopefully) ## Comment: With these changes, several problems in tianshou's inheritance hierarchy become more apparent. I tried highlighting them for future work. --------- Co-authored-by: Dominik Jain <d.jain@appliedai.de>
95 lines
3.5 KiB
Python
95 lines
3.5 KiB
Python
from collections.abc import Callable
|
|
from typing import Any
|
|
|
|
from tensorboard.backend.event_processing import event_accumulator
|
|
from torch.utils.tensorboard import SummaryWriter
|
|
|
|
from tianshou.utils.logger.base import LOG_DATA_TYPE, BaseLogger
|
|
from tianshou.utils.warning import deprecation
|
|
|
|
|
|
class TensorboardLogger(BaseLogger):
|
|
"""A logger that relies on tensorboard SummaryWriter by default to visualize and log statistics.
|
|
|
|
:param SummaryWriter writer: the writer to log data.
|
|
:param train_interval: the log interval in log_train_data(). Default to 1000.
|
|
:param test_interval: the log interval in log_test_data(). Default to 1.
|
|
:param update_interval: the log interval in log_update_data(). Default to 1000.
|
|
:param save_interval: the save interval in save_data(). Default to 1 (save at
|
|
the end of each epoch).
|
|
:param write_flush: whether to flush tensorboard result after each
|
|
add_scalar operation. Default to True.
|
|
"""
|
|
|
|
def __init__(
|
|
self,
|
|
writer: SummaryWriter,
|
|
train_interval: int = 1000,
|
|
test_interval: int = 1,
|
|
update_interval: int = 1000,
|
|
save_interval: int = 1,
|
|
write_flush: bool = True,
|
|
) -> None:
|
|
super().__init__(train_interval, test_interval, update_interval)
|
|
self.save_interval = save_interval
|
|
self.write_flush = write_flush
|
|
self.last_save_step = -1
|
|
self.writer = writer
|
|
|
|
def write(self, step_type: str, step: int, data: LOG_DATA_TYPE) -> None:
|
|
for k, v in data.items():
|
|
self.writer.add_scalar(k, v, global_step=step)
|
|
if self.write_flush: # issue 580
|
|
self.writer.flush() # issue #482
|
|
|
|
def save_data(
|
|
self,
|
|
epoch: int,
|
|
env_step: int,
|
|
gradient_step: int,
|
|
save_checkpoint_fn: Callable[[int, int, int], str] | None = None,
|
|
) -> None:
|
|
if save_checkpoint_fn and epoch - self.last_save_step >= self.save_interval:
|
|
self.last_save_step = epoch
|
|
save_checkpoint_fn(epoch, env_step, gradient_step)
|
|
self.write("save/epoch", epoch, {"save/epoch": epoch})
|
|
self.write("save/env_step", env_step, {"save/env_step": env_step})
|
|
self.write(
|
|
"save/gradient_step",
|
|
gradient_step,
|
|
{"save/gradient_step": gradient_step},
|
|
)
|
|
|
|
def restore_data(self) -> tuple[int, int, int]:
|
|
ea = event_accumulator.EventAccumulator(self.writer.log_dir)
|
|
ea.Reload()
|
|
|
|
try: # epoch / gradient_step
|
|
epoch = ea.scalars.Items("save/epoch")[-1].step
|
|
self.last_save_step = self.last_log_test_step = epoch
|
|
gradient_step = ea.scalars.Items("save/gradient_step")[-1].step
|
|
self.last_log_update_step = gradient_step
|
|
except KeyError:
|
|
epoch, gradient_step = 0, 0
|
|
try: # offline trainer doesn't have env_step
|
|
env_step = ea.scalars.Items("save/env_step")[-1].step
|
|
self.last_log_train_step = env_step
|
|
except KeyError:
|
|
env_step = 0
|
|
|
|
return epoch, env_step, gradient_step
|
|
|
|
|
|
class BasicLogger(TensorboardLogger):
|
|
"""BasicLogger has changed its name to TensorboardLogger in #427.
|
|
|
|
This class is for compatibility.
|
|
"""
|
|
|
|
def __init__(self, *args: Any, **kwargs: Any) -> None:
|
|
deprecation(
|
|
"Class BasicLogger is marked as deprecated and will be removed soon. "
|
|
"Please use TensorboardLogger instead.",
|
|
)
|
|
super().__init__(*args, **kwargs)
|