What is the default run naming convention in MLflow?

In MLflow, the default run naming convention is designed to provide unique and informative identifiers for each experiment run. When you initiate an MLflow run without specifying a custom name, MLflow automatically generates a name for it.

By default, MLflow run names follow a pattern that combines a few key elements:

1. Adjective: MLflow starts by selecting a random adjective from a predefined list. This adds a more human-friendly element to the naming convention.

2. Noun: Next, it appends a random noun, also selected from a predefined list. This further enhances the uniqueness and readability of the run name.

3. Unique ID: Finally, a unique identifier is added to guarantee that each run name is distinct, even if the chosen adjective and noun combination is repeated. This identifier is often a short alphanumeric string.

Therefore, the default run name will look something like "whimsical-flower-8b4a", where whimsical is the adjective, flower is the noun, and 8b4a is the unique identifier.

This default convention is helpful for:

- Uniqueness: Ensuring each run has a distinct name.

- Readability: Providing somewhat descriptive names at a glance.

- Ease of Use: Eliminating the need to manually name every run, especially when doing quick experiments or when a specific naming scheme isn't essential.

However, MLflow also allows you to specify custom run names if the default naming is not suitable for your particular needs. You can provide a run name during the start of your run, for example by:

with mlflow.start_run(run_name="My_Custom_Run"):
   # Your MLflow code here
pass

This will name your run as "My_Custom_Run" instead of the auto-generated one.

In conclusion, while MLflow's default run naming convention is convenient for getting started, you have full control over how your runs are named and can use custom names when needed.

Learn more about this