REPRODUCE_RESULTS.md Unclear about data paths

[carbonox-infernox]

In REPRODUCE_RESULTS.md it says:

set paths in neptune.yaml

data_dir: /path/to/data meta_dir: /path/to/data masks_overlayed_prefix: masks_overlayed experiment_dir: /path/to/work/dir

However, it doesn't say what these directories correspond to from the downloaded dataset. For example, I could assume that data_dir is supposed to be the train folder, but what about meta_dir and experiment_dir?

[taraspiotr]

Hi! data_dir is a directory of the dataset. meta_dir is a directory in which metadata.csv file will be saved after running method prepare_metadata. You can set it to wherever you want, the file with metadata will be saved there and later read from there. experiment_dir is a directory in which all the files of an experiment will be saved, so training checkpoints, trained model, submission etc. You cant set is however you want as well.

[carbonox-infernox]

Thanks! I think reproduce_results would be improved by the inclusion of this.

[carbonox-infernox]

I'm reopening this issue, because the only way to make this work is to set data_dir: /input/data in neptune.yaml. I think reproduce_results.md should be updated to explain this for two reasons:

1. reproduce_results.md currently makes no mention of an input folder, and
2. There is no input folder visible from within Neptune

Therefore using input/data for the path is completely unguessable. I was stuck on this for a long time as a result of that: #182