How to Log Artifacts in MLflow: Simple Guide with Examples

To log artifacts in MLflow, use mlflow.log_artifact() to save a single file or mlflow.log_artifacts() to save all files in a directory. These functions store files like models, images, or logs in the current MLflow run's artifact storage.

Syntax

The main functions to log artifacts in MLflow are:

• mlflow.log_artifact(local_path, artifact_path=None): Logs a single file from local_path. Optionally, you can specify artifact_path to organize artifacts in folders.
• mlflow.log_artifacts(local_dir, artifact_path=None): Logs all files from the directory local_dir. You can also specify artifact_path to place them inside a folder.

Both functions must be called inside an active MLflow run context.

python

import mlflow

# Log a single file
mlflow.log_artifact(local_path='path/to/file.txt', artifact_path='my_folder')

# Log all files in a directory
mlflow.log_artifacts(local_dir='path/to/folder', artifact_path='my_folder')

Example

This example shows how to create a text file and log it as an artifact in MLflow. It demonstrates starting a run, logging the artifact, and ending the run.

python

import mlflow

# Create a sample file to log
file_name = 'example.txt'
with open(file_name, 'w') as f:
    f.write('Hello, MLflow artifacts!')

# Start an MLflow run
with mlflow.start_run():
    # Log the file as an artifact
    mlflow.log_artifact(file_name, artifact_path='texts')
    print(f'Logged artifact: {file_name}')

Output

Logged artifact: example.txt

Common Pitfalls

• Not starting an MLflow run: You must call mlflow.start_run() before logging artifacts, or you will get an error.
• Incorrect file paths: Ensure the file or directory exists at the given path before logging.
• Confusing log_artifact and log_artifacts: Use log_artifact for single files and log_artifacts for directories.

python

import mlflow

# Wrong: logging without a run
try:
    mlflow.log_artifact('example.txt')
except Exception as e:
    print(f'Error: {e}')

# Right: logging inside a run
with mlflow.start_run():
    mlflow.log_artifact('example.txt')
    print('Artifact logged successfully')

Output

Error: No active run. mlflow.start_run() should be used to start a run.
Artifact logged successfully

Quick Reference

Function	Purpose	Parameters
mlflow.log_artifact	Log a single file as an artifact	local_path: file path, artifact_path: optional folder
mlflow.log_artifacts	Log all files in a directory	local_dir: directory path, artifact_path: optional folder
mlflow.start_run	Start an MLflow run context	No parameters needed
mlflow.end_run	End the current MLflow run	No parameters needed

Key Takeaways

Always call mlflow.start_run() before logging artifacts to create an active run.

Use mlflow.log_artifact() for single files and mlflow.log_artifacts() for directories.

Specify artifact_path to organize artifacts inside folders in the run's storage.

Check that file or directory paths exist before logging to avoid errors.

Artifacts are stored with the run and can be viewed or downloaded later from the MLflow UI.