Please teach me how to modify the code

Hey @ashwinvaidya17, Is this issue been worked on?
I am interested to work on this, can you please assign it to me?

Hi @ChaitanyaYeole02, no one is working on this at the moment. Thanks for your interest

Hey @samet-akcay, what kind of metric could be written in this CSV file?
As I saw for the other logging files, images are been logged.
Currently, I have written the basic class to log the metrics.

class AnomalibCSVLogger(CSVLogger):
    def __init__(
            self,
            save_dir: str,
            name: str | None = "default",
            version: int | str | None = None,
            prefix: str = "",
            flush_logs_every_n_step: int | None = 100,
    ) -> None:
        super().__init__(
            save_dir=save_dir,
            name=name,
            version=version,
            prefix=prefix,
            flush_logs_every_n_steps=flush_logs_every_n_step,
        )

Hey @samet-akcay, any suggestions?

@ChaitanyaYeole02 I tried your code locally. This is what it writes

Looks fine to me. Feel free to create a PR.

MRE

from anomalib.data import MVTec
from anomalib.engine import Engine
from anomalib.loggers.csv import CSVLogger
from anomalib.models import Padim


def main():
    logger = CSVLogger(save_dir="csv_test")
    engine = Engine(logger=logger)
    model = Padim()
    datamodule = MVTec()

    engine.train(datamodule=datamodule, model=model)


if __name__ == "__main__":
    main()

@ashwinvaidya17 I will raise the PR, but could you also tell me how can I setup and test this locally?

Currently, the subclassed logger provides the same features as the base logger. Do you plan to save separate test and train metrics? Epoch 0 and 1 will be a bit confusing. I tested this on Padim so ideally, we should get train metrics, validation metrics, and test metrics separately rather than showing up as epoch. Otherwise this does not provide any additional features on top of the base class.

Yeah, currently it does not provide any feature on top of the base class. Do you have any suggestion of additional features on top of the base class?

I was able to replicate the MRE. Should we add more metrics?

@ChaitanyaYeole02 thanks for creating the PR, but apologies, I realized that my issue is ill-defined. The original use-case came from the benchmarking script as it prints all the models, dataset categories, and metrics in a single table. Something like this

I was hoping we could get something similar for general training but looks like the information is not sufficient. With a closer look at the base class, it seems like we can only access the step along with the metrics dict. This makes it harder to identify which stage of training it is printed from. Ideally, it would be nice to save the metrics like

But looks like it is not as straightforward as I thought.

Ohh, interesting. I would take a look at this. How did you get these benchmark results any MRE?

https://anomalib.readthedocs.io/en/latest/markdown/guides/reference/pipelines/benchmark/index.html

Thank you, let me take a look at this and if I am stuck somewhere I will let you know.