[Question] Using tensorboard with custom logger

[JDRanpariya]

❓ Question

How can we make it work through custom logger? I am using custom logger for custom callback and do want to have my logs show up in wandb using tensorboard.

I saw it being mentioned in #1997 but that doesn't help much.

class CustomLogger(Logger):
    def __init__(self, folder: Optional[str], output_formats: list[KVWriter]):
        super(CustomLogger, self).__init__(folder, output_formats)

    def dump(self, step: int = 0) -> None:
        """
        Write all of the diagnostics from the current iteration without clearing the stored values.
        """
        if self.level == DISABLED:
            return
        for _format in self.output_formats:
            if isinstance(_format, KVWriter):
                _format.write(self.name_to_value, self.name_to_excluded, step)

Checklist

• I have checked that there is no similar issue in the repo
• I have read the documentation
• If code there is, it is minimal and working
• If code there is, it is formatted using the markdown code blocks for both code and stack traces.

[araffin]

Hello,
what is you current problem?
what are you trying to log and what works/doesn't work?

[JDRanpariya]

It's actually logging everything that's on https://stable-baselines3.readthedocs.io/en/master/common/logger.html, I wanted to access the eval/, rollout/ and time/ in a callback. So I used custom logger and removed the line that was dumping that info.

But on using inherited Logger class, it overwrites tensorboard_log and verbose settings passed to the constructor, I want to enable tensorflow_log with custom logger as I want to log the values to wandb along with having access to eva/, rollout and time/.

I hope that is clear enough, if not please let me know.

[araffin]

from pathlib import Path
from stable_baselines3 import A2C
from stable_baselines3.common.logger import Logger, make_output_format

# Tensorboard log dir
tmp_path = Path("/tmp/sb3_log/")
tmp_path.mkdir(exist_ok=True)
folder = str(tmp_path)

# set up logger
class CustomLogger(Logger):
    def dump(self, step: int = 0) -> None:
        print(self.name_to_value)
        super().dump(step)

format_strings = ["stdout", "tensorboard"]
output_formats = [make_output_format(f, folder) for f in format_strings]
logger = CustomLogger(folder=folder, output_formats=output_formats)

model = A2C("MlpPolicy", "CartPole-v1", verbose=1)
# Set new logger
model.set_logger(logger)
model.learn(5_000)

?

[JDRanpariya]

Thanks! It's working, I wasn't aware of make_out_format.