Saving and loading a model

With the Model API, models can be shared between developers who may not have access to the same MLflow Tracking server by using a local filesystem.

In this exercise, you will train a new LinearRegression model from an existing one using the Unicorn dataset. First, you will load an existing model from the local filesystem. Then you will train a new model from the existing model and save it back to the local filesystem.

The existing model has been saved to the local filesystem in a directory called "lr_local_v1". The mlflow module will be imported.

This exercise is part of the course

Introduction to MLflow

Exercise instructions

Load the model from the local filesystem directory "lr_local_v1" using scikit-learn library from the MLflow module.
Using the scikit-learn library from the mlflow module, save the model locally to a directory called "lr_local_v2".

Hands-on interactive exercise

Have a go at this exercise by completing this sample code.

# Load model from local filesystem
model = ____.____.____("____")

# Training Data
X = df[["R&D Spend", "Administration", "Marketing Spend", "State"]]
y = df[["Profit"]]
X_train, X_test, y_train, y_test = train_test_split(X, y, train_size=0.7,random_state=0)
# Train Model
model.fit(X_train, y_train)

# Save model to local filesystem
____.____.____(____, "____")

Edit and Run Code

This exercise is part of the course

Introduction to MLflow

AdvancedSkill Level

4.8+

Start Course for Free

In this Chapter, you will be introduced to MLflow and how it aims to assist with some difficulties of the Machine Learning lifecycle. You will be introduced to the four main concepts that make up MLflow with a main focus on MLflow Tracking. You will learn to create experiments and runs as well as how to track metrics, parameters, and artifacts. Finally, you will search MLflow programmatically to find experiment runs that fit certain criteria.

Exercise 1: What is MLflow?Exercise 2: Components of MLflow Exercise 3: MLflow experiments Exercise 4: MLflow Tracking Exercise 5: Starting a run Exercise 6: Logging a run Exercise 7: How to retrieve active run data?Exercise 8: Querying runs Exercise 9: Search runs query options Exercise 10: Searching runs

In this Chapter, you will be introduced to MLflow Models. The MLflow Models component of MLflow plays an essential role in the Model Evaluation and Model Engineering steps of the Machine Learning lifecycle. You will learn how MLflow Models standardizes the packaging of ML models as well as how to save, log and load them. You will learn how to create custom MLflow Models to provide more flexibility to your use cases as well as how to evaluate model performance. You will utilize the powerful concept of “Flavors” and finally use the MLflow command line tool for model deployment.

Exercise 1: Introduction to MLflow Models Exercise 2: Package a machine learning model Exercise 3: Storage Format Exercise 4: What's in an MLmodel file?Exercise 5: Model API Exercise 6: Saving and loading a model

Current Exercise

Exercise 7: Logging and loading a model Exercise 8: Custom models Exercise 9: Creating a custom Python Class Exercise 10: Custom scikit-learn model Exercise 11: Scikit-learn flavor and evaluation Exercise 12: Model serving Exercise 13: Serving a model Exercise 14: Score from a served model

This Chapter introduces the concept of MLflow called the Model Registry. You will be introduced to how the Model Registry is used to manage the lifecycle of ML models. You will learn how to create and search for models in the Model Registry. You then learn how to register models to the Model Registry and learn how to transition models between predefined stages. Finally, you will also learn how to deploy models from the Model Registry.

Exercise 1: Introduction to MLflow Model Registry Exercise 2: Create a Model Exercise 3: Searching Models Exercise 4: Registering Models Exercise 5: Registering existing models Exercise 6: Registering new models Exercise 7: Model stages Exercise 8: Model stages in Model Registry Exercise 9: Transitioning model stages Exercise 10: Model deployment Exercise 11: Serving models from the Model Registry Exercise 12: Loading models from the Model Registry

In this chapter, you'll gain valuable knowledge on how to streamline your data science code for reusability and reproducibility using MLflow Projects. The chapter begins by introducing the concept of MLflow Projects and walking you through creating an MLproject file. From there, you'll learn how to run MLflow Projects through both the command-line and the MLflow Projects module while also discovering the power of using parameters for added flexibility in your code. Finally, you will learn how to manage steps of the machine learning lifecycle by creating a multi-step workflow using MLflow Projects.

Exercise 1: Introduction to MLflow Projects Exercise 2: MLproject file layout Exercise 3: Creating an MLproject Exercise 4: Running MLflow Projects Exercise 5: MLflow run command Exercise 6: MLflow projects module Exercise 7: Specifying parameters Exercise 8: Adding parameters to MLproject Exercise 9: Adding parameters to project run Exercise 10: Workflows Exercise 11: Creating an MLproject for the ML Lifecycle: Model Engineering Exercise 12: Creating an MLproject for the ML Lifecycle: Model Evaluation Exercise 13: Creating a multi-step workflow: Model Engineering Exercise 14: Creating a multi-step workflow: Model Evaluation Exercise 15: Wrap-up