Open kongzii opened 2 years ago
Hi @amesar, sorry to bother you, but do you have any feedback for this PR, please? Is there something I should change?
Hi @amesar, any feedback for this, please? If you don't want to merge this, no problem, I can fork the project and do the required modifications just for me, but I thought it would be a shame to not collaborate on a project like this.
Hi, thank you for this project. I am working on a simple script that will use this to migrate runs and models between our development and production mlflow server in the company, however, I found a couple of issues so I tried to fix them.
mlruns
directory on import of this library, e.g. doingfrom mlflow_export_import.run import export_run, import_run
would create a new local mlflow server, if the MLFLOW_TRACKING_URI variable is not set.host
argument to the initializators, it is passed to the clients that the class is using, the reason is that before, you was able to pass custom mlflow client, but not the host, so the host could be different than tracking URI of the client.if not dst_source.startswith("dbfs:") and not dst_source.startswith("s3:") and not os.path.exists(dst_source)
, because it was failing on S3 path, saying it doesn't exist.mlflow.set_experiment
, because it was setting a global experiment, not for the particular client.The result is that now, you are able to use this library from the other python scripts, without need of covering environment variables, example:
I will do probably more contributions in the future, but I wanted to get your feedback before there is too many of changes.