Custom models

Introduction to MLflow

Weston Bassler

Senior MLOps Engineer

Example use cases

  • NLP - Tokenizer(s)

  • Classification - Label encoder

  • Pre/Post processing

  • Not a built-in flavor

Different

1 unsplash.com
Introduction to MLflow

Custom Python models

  • Built in Flavor - python_function

  • mlflow.pyfunc

    • save_model()
    • log_model()
    • load_model()
Introduction to MLflow

Custom model class

  • Custom model class

    • MyClass(mlflow.pyfunc.PythonModel)
  • PythonModel class

    • load_context() - loads artifacts when mlflow.pyfunc.load_model() is called
    • predict() - takes model input and performs user defined evaluation
Introduction to MLflow

Python class

# Class
class MyPythonClass:

# Function that prints Hello! def my_func(): print("Hello!")
# Create a new Object
x = MyPythonClass()

# Excute my_func function x.my_func()
"Hello!"
Introduction to MLflow

Example custom Class

import mlflow.pyfunc

# Define the model class
class CustomPredict(mlflow.pyfunc.PythonModel):

# Load artifacts def load_context(self, context): self.model = mlflow.sklearn.load_model(context.artifacts["custom_model"])
# Evaluate input using custom_function() def predict(self, context, model_input): prediction = self.model.predict(model_input) return custom_function(prediction)
Introduction to MLflow

Saving and logging a custom model

# Save model to local filesystem
mlflow.pyfunc.save_model(path="custom_model", python_model=CustomPredict())
# Log model to MLflow Tracking
mlflow.pyfunc.log_model(artifact_path="custom_model", python_model=CustomPredict())
Introduction to MLflow

Loading custom models

# Load model from local filesystem
mlflow.pyfunc.load_model("local")
# Load model from MLflow Tracking
mlflow.pyfunc.load_model("runs:/run_id/tracking_path")
Introduction to MLflow

Model Evaluation

  • mlflow.evaluate() - Performance based on a dataset

  • Regression and Classification models

Introduction to MLflow

Evaluation Example

# Training Data
X_train, X_test, y_train, y_test = \
    train_test_split(X, y,
    train_size=0.7,random_state=0)

# Linear Regression model lr = LinearRegression() lr.fit(X_train, y_train)
# Dataset
eval_data = X_test
eval_data["test_label"] = y_test

# Evaluate model with Dataset mlflow.evaluate( "runs:/run_id/model", eval_data, targets="test_label", model_type="regressor" )
Introduction to MLflow

Tracking UI

eval

shap

1 shap.readthedocs.io
Introduction to MLflow

Let's practice!

Introduction to MLflow

Preparing Video For Download...