Skip to content

Experiment Tracking Logging

import practicuscore as prt
import os
import mlflow

Code Explanation

This code sets up and configures a Practicus AI experiment service using the practicuscore library. It connects to an external experiment tracking service, such as MLflow, and optionally creates a new experiment.

Code Breakdown

  1. import practicuscore as prt

  2. Imports the Practicus AI core library for managing experiments and deployments.

  3. import os

  4. Standard library module for interacting with the operating system (not used directly in this snippet).

  5. import mlflow

  6. Imports the MLflow library, which is commonly used for experiment tracking.

  7. service_name = "My Mlflow Service"

  8. Sets the name of the experiment tracking service. For example, MLflow is identified here as "My Mlflow Service".

  9. service_key = ""

  10. A secure key required to authenticate with the Practicus AI service. This key can be obtained from the Practicus AI Admin Console.

  11. experiment_name = None

  12. (Optional) Specifies a name for a new experiment. If None, no new experiment is created during configuration.

  13. prt.experiments.configure()

  14. Purpose: Configures the experiment service with the provided parameters.
  15. Parameters:
    • service_name: The name of the service being configured.
    • service_key: Authentication key for accessing the service.
    • experiment_name: Optional name for a new experiment.
service_name = "My Mlflow Service"
# You need to configure using the service unique key, you can find your key on the "Practicus AI Admin Console" 
service_key =  ""
# Optionally, you can provide experiment name to create a new experiment while configuring
experiment_name = None

prt.experiments.configure(service_name=service_name, service_key=service_key, experiment_name=experiment_name)

# Set experiment name, if you haven't already while configuring the service
mlflow.set_experiment("my experiment")

# Prefer unique run names, or leave empty to auto generate unique names
run_name = "My ML experiment run 123"

# Start an MLflow run and log params, metrics and artifacts
with mlflow.start_run(run_name=run_name):
    # Log parameters
    mlflow.log_param("param1", 5)
    mlflow.log_param("param2", "test")

    # Log metrics
    mlflow.log_metric("metric1", 0.85)

    # Create an artifact (e.g., a text file)
    artifact_path = "artifacts"
    if not os.path.exists(artifact_path):
        os.makedirs(artifact_path)
    file_path = os.path.join(artifact_path, "output.txt")

    with open(file_path, "w") as f:
        f.write("This is a test artifact.")

    # Log the artifact
    mlflow.log_artifacts(artifact_path)

    # Optional: Print the run ID
    print("Run ID:", mlflow.active_run().info.run_id)

# Explicitly close the active MLflow run, if you are not using the above with keyword
# mlflow.end_run()

Previous: SparkML Ice Cream | Next: Experiment Tracking Model Training