mlflow.md

Logging Experiments with MLFlow

Starting the mlflow tracking server

In a terminal that has the virtual environment active, start the mlflow server.

mlflow server --host 127.0.0.1 --port 8080

Alternatively, you can run the "Start local MLflow server" task in vscode by pressing Ctrl+Shift+P, searching for "Run Task", then search for "Start local MLflow server". This will create a new terminal inside of VSCode running the local server.

Note: By default, the server creates mlartifacts/ and mlruns/ in the current directory for storing data about runs.

Logging a run and viewing the results

Now run the inference script. In the console output of the script, MLFlow will print something like "View experiment at ". If you open that URL in a browser window, you should see a list of the "runs" of this "experiment". Click on the most recent one to see the results of this particular run.

There are a few tabs near the top of the page for the run. On the "Overview" tab you can see the parameters used for this run. On the "System metrics" tab you can see the GPU memory, CPU, etc. used at various points during the run. On the "Artifacts" tab you can see the files created during the run (for example the annotations created during an inference run.)

Disabling mlflow logging

Sometimes you may want to run your code without the mlflow server running. To do that, unset the MLFLOW_TRACKING_URI environment parameter. This actually sets the tracking uri to file:///tmp/mlruns so that logging happens to a temporary location on disk instead of the mlflow tracking server.