Skip to content

🧪 MLflow Integration

What you'll learn

How to auto-log metrics and models to MLflow — FlowyML + MLflow = automated experiment tracking with the industry-standard platform.

Track experiments, manage model versions, and deploy models using MLflow's open-source ecosystem.

Why MLflow?

Feature Benefit
Experiment Tracking Log parameters, metrics, and artifacts
Model Registry Version and manage model lifecycles
Universal Works with any ML library
Open Source No vendor lock-in

🧪 Setup

1
2
pip install mlflow
mlflow server --host 127.0.0.1 --port 5000  # Start tracking server

Auto-Logging

Enable MLflow tracking for your pipeline:

1
2
3
4
5
6
7
8
from flowyml.integrations.mlflow import MLflowTracker

pipeline.run(
    tracker=MLflowTracker(
        tracking_uri="http://localhost:5000",
        experiment_name="my_experiment",
    )
)

MLflowTracker Parameters

Parameter Type Default Description
tracking_uri str required MLflow tracking server URL
experiment_name str "default" Experiment name
run_name str None Custom run name
auto_log bool True Auto-log step params/metrics

📝 Custom Logging in Steps

Log custom metrics, parameters, and artifacts:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
import mlflow
from flowyml import step

@step
def train_model(data):
    # Log parameters
    mlflow.log_param("lr", 0.01)
    mlflow.log_param("optimizer", "adam")

    model = train(data, lr=0.01)

    # Log metrics
    mlflow.log_metric("accuracy", 0.95)
    mlflow.log_metric("f1_score", 0.93)

    # Log the model artifact
    mlflow.sklearn.log_model(model, "model")

    return model

📦 Model Registry

Register and promote models through lifecycle stages:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
@step
def register_best_model(model, metrics):
    # Log and register in one step
    mlflow.sklearn.log_model(model, "model", registered_model_name="ProductionModel")

    # Promote to staging
    client = mlflow.tracking.MlflowClient()
    client.transition_model_version_stage(
        name="ProductionModel",
        version="1",
        stage="Staging",
    )

Best Practices

Use autolog for quick wins

mlflow.autolog() automatically captures sklearn, XGBoost, LightGBM, and PyTorch metrics — zero code changes needed.

Remote tracking server

In production, point tracking_uri to a shared MLflow server so your whole team can see experiment results.