About

aimlflow helps to explore various types of metadata tracked during the training with MLFLow, including:

• hyper-parameters
• metrics
• images
• audio
• text

More about Aim: https://github.com/aimhubio/aim

More about MLFLow: https://github.com/mlflow/mlflow

Getting Started

Follow the steps below to set up aimlflow.

1. Install aimlflow on your training environment:

pip install aim-mlflow

2. Run live time convertor to sync MLFlow logs with Aim:

aimlflow sync --mlflow-tracking-uri={mlflow_uri} --aim-repo={aim_repo_path}

3. Run the Aim UI:

aim up --repo={aim_repo_path}

Why use aimlflow?

1. Powerful pythonic search to select the runs you want to analyze.

2. Group metrics by hyperparameters to analyze hyperparameters’ influence on run performance.

3. Select multiple metrics and analyze them side by side.

4. Aggregate metrics by std.dev, std.err, conf.interval.

5. Align x axis by any other metric.

6. Scatter plots to learn correlations and trends.

7. High dimensional data visualization via parallel coordinate plot.