Galileo-Galilei/kedro-mlflow-tutorial

run_id in MLflow tag is a timestamp and not an ID

Chouffe opened this issue · 1 comments

After following the tutorial and inspecting the results via the MLflow UI, the run_id set as a tag is a timestamp and not an id.
I am wondering what is wrong here.

image

Hi,

it may sounds confusing but all the tags set in mlflow come from Kedro Journal. This is an object of the kedro==0.16.x versions which tracks some metadata about your run (including the parameters passed to CLI), and the internal Kedro's (not mlflow) run_id which is just a timestamp. This timestamp is useful in case you use Kedro's native versioning abilities for datasets.

You should really not care too much about it for two reasons:

  • it is not very informative: you already know the start date of your mlflow run
  • it is deprecated in kedro==0.17.x and will be removed in kedro==0.18.x.
    I plan to update the example to use a more recent version of kedro (I initially used kedro 0.16.x because when I created this repo 0.17.x was in early stage and have some really annoying bugs, but kedro 0.17.4 seems stable.