Where MLflow runs are logged

All MLflow runs are logged to the active experiment, which can be set using any of the following ways:

• Use the mlflow.set_experiment() command.
• Use the experiment_id parameter in the mlflow.start_run() command.
• Set one of the MLflow environment variables MLFLOW_EXPERIMENT_NAME or MLFLOW_EXPERIMENT_ID.

If no active experiment is set, runs are logged to the notebook experiment.

MLflow Logging

# basic run
with mlflow.start_run(run_name="Basic RF run"):
  
  # Set the model parameters. 
  n_estimators = 100
  max_depth = 6
  max_features = 3
  
  # Create and train model.
  rf = RandomForestRegressor(n_estimators = n_estimators, max_depth = max_depth, max_features = max_features)
  rf.fit(X_train, y_train)
  
  # Use the model to make predictions on the test dataset.
  predictions = rf.predict(X_test)
  
  # Log the model parameters used for this run.
  mlflow.log_param("num_trees", n_estimators)
  mlflow.log_param("maxdepth", max_depth)
  mlflow.log_param("max_feat", max_features)
  
  # Define a metric to use to evaluate the model.
  mse = mean_squared_error(y_test, predictions)
    
  # Log the value of the metric from this run.
  mlflow.log_metric("mse", mse)
  
  # Log the model created by this run.
  mlflow.sklearn.log_model(rf, "random-forest-model")
  # mlflow.pyfunc.log_model("name_of_model", python_model=model, env) 
  
  # Save the table of predicted values
  savetxt('predictions.csv', predictions, delimiter=',')
  
  # Log the saved table as an artifact
  mlflow.log_artifact("predictions.csv")

 

MLflow Experiment Access

expId = mlflow.get_experiment_by_name("/Shared/diabetes_experiment/").experiment_id
df = spark.read.format("mlflow-experiment").load(expId)
display(df)

# OR
df = spark.read.format("mlflow-experiment").load("3270527066281272")
filtered_df = df.filter("metrics.loss < 0.01 AND params.learning_rate > '0.001'")
display(filtered_df)

from mlflow.store.artifact.models_artifact_repo import ModelsArtifactRepository

model_uri = MlflowClient.get_model_version_download_uri(model_name, model_version)

from mlflow.tracking import MlflowClient

client = MlflowClient()

display(clienet.list_run_infos(experiment_id))

runs = spark.read.format("mlflow-experiment").load(experiment_id)

run_rf = runs.orderBy("start_time", ascending=False).first()

client.list_artifacts(run_rf.run_id)

client.get_run(run_rf.run_id).data.metrics

Advanced Experimentation

<aside> 💡

Background: https://mlflow.org/docs/latest/model/signatures.html#mlflow-model-signatures-and-input-examples-guide

</aside>

Signatures and Input Examples

Previously, when logging a model in MLflow we only logged the model and name for the model artifact with .log_model(model, model_name).However, it is a best practice to also log a model signature and input example. This allows for better schema checks and, therefore, integration with automated deployment tools.