7

I want to run the mlflow ui from a different folder. By default it creates a folder called 'mlruns' in the folder of my user. If create runs in a Jupiter-Notebook using a specific working directory a new Folder 'mlruns' is created in that directory. My goal is to run the mlflow ui acccessing the 'mlruns' Folder in this self specified Directory. So far I have tried:

running mlflow ui in the anaconda prompt PowerShell from the Directory i use

defining the Directory in running the mlflow Server by:

mlflow ui --backend-store-uri file:///'directory here'

I use Windows, Anaconda, Python, Jupiter-Notebook and the latest mlflow Version (1.10)

MatthiasHerp
  • 271
  • 3
  • 9

1 Answers1

11

This worked for me after some help on Github:

Starting the Mlflow server for a certain location (type in the anconda prompt powershell):

Here the folder in my home directory of the user mah mlflow ui

mlflow ui --backend-store-uri file:///Users/mah/mlruns

Setting the place where to store runs (type in python):

For the first folder above:

mlflow.set_tracking_uri("file:///Users/mah/mlruns")
Sida Zhou
  • 3,529
  • 2
  • 33
  • 48
MatthiasHerp
  • 271
  • 3
  • 9
  • 2
    I agree with the answer. The path for mlflow uri needs to be a full path. If you delare default-artifact-uri=‘mlflow’ it will create a local folder in your Jupy Notebook folder. If you declare it as ‘file:///FullPath/‘ it will work fine – pushd93 Oct 08 '20 at 15:04
  • A relative path works too, e.g., `mlflow ui --backend-store-uri notebooks/mlruns` – SergiyKolesnikov Aug 07 '23 at 17:08