LoggerCollection name redundancy

[twsl]

🐛 Bug

When you create a logger collection, the name of the logger collection is used for the save path in ModelCheckpoint. LoggerCollection concats the names of all logger. Therefore, in my example the checkpoints are saved to f"{experiment_name}_{experiment_name}" instead of experiment_name.

To Reproduce

tb_logger = TensorBoardLogger(save_dir=root_dir, name=experiment_name, version=version)
csv_logger = CSVLogger(save_dir=root_dir, name=experiment_name, version=version)
trainer = pl.Trainer(
        logger=[
            csv_logger,
            tb_logger,
        ],
    )

#...

class TestModule(pl.LightningModule):
    def configure_callbacks(self):
        checkpoint = ModelCheckpoint(
            monitor="val_map",
            filename="{epoch}-{step}--{val_map:.4f}",
            auto_insert_metric_name=True,
            every_n_epochs=1,
        )
        return [checkpoint]

Expected behavior

I would expect the checkpoints to be saved under the same name I set for the loggers. I understand the motive behind the current implementation, but I think only unique names should be concatenated there. Same goes for the version property.

Environment

• PyTorch Lightning Version (e.g., 1.5.0): 1.5.4
• PyTorch Version (e.g., 1.10): 1.10
• Python version (e.g., 3.9): 3.8
• OS (e.g., Linux): Linux
• CUDA/cuDNN version:
• GPU models and configuration: 1080
• How you installed PyTorch (conda, pip, source): pip
• If compiling from source, the output of torch.__config__.show(): -
• Any other relevant information:

Additional context

Workaround is to set the dirpath argument manually.

cc @Borda @awaelchli @edward-io @ananthsub @tchaton

[tchaton]

Dear @twsl,

Would you be interested in contributing a fix?

Best,
T.C

[ashutoshml]

I can help with fixing this issue.

[tchaton]

Great @ashutoshml, I assigned you to the issue.

[ashutoshml]

Thanks. I'll start working on it.

What are the requirements for the fix, though?

Based on the issue reported, I assume the fix should only be concerning experiment_name in the loggers. Is there anything else that needs to change?

[tchaton]

Hey @ashutoshml,

I would assume this to fix the problem. There might be some edge case if the users wanted to have twice of the same type of loggers though.

[ashutoshml]

Sure. I can take a look at that.
Currently the issue is assigned to you, though.