Saving and Loading an Ax Experiment to JSON

Ax provides a convenient way to save and load experiments in JSON format, making it easy to store and transport experiment data. In this recipe, we will walk through the steps of saving and loading an Ax experiment to JSON using the Client.

Introduction

Saving and loading an experiment to JSON is useful when you want to store the experiment data in a lightweight and transportable format. This can be particularly useful for users who prefer a simple storage solution or need to share experiment data with others.

Steps

In this recipe, we plan to save a a snapshot of an Ax Client to JSON, reload the client from the JSON file, and validate its contents.

• Saving to JSON
  1. Initialize a Client and configure it with an experiment
  2. View the state of the attached experiment via summarize()
  3. Call save_to_json_file to save a snapshot of the Client to JSON
• Loading from JSON
  1. Call load_from_json_file to initialize a new Client
  2. Validate the state of the attached experiment via summarize()

Saving to JSON

1. Initialize a Client and configure it with an experiment

Instantiate a Client and configure it for your experiment.

client = Client()

client.configure_experiment(...)
client.configure_optimization(...)

2. View the state of the attached experiment via summarize()

You can inspect the state of a experiment by leveraging the summarize() method, which returns a DataFrame

client.summarize()

2. Call save_to_json_file to save a snapshot of the Client to JSON

In order to save an experiment to JSON, we need to call the save_to_json_file method on the Client instance. This method takes a single optional argument filepath, which is the filepath where we want to save the JSON file (argument defaults to "ax_client_snapshot.json").

client.save_to_json_file()

On success, this will save a snapshot of the Client's settings and state to the specified file, which includes information about the experiment and generation strategy (if present).

Load an Experiment from JSON

1. Call load_from_json_file to initialize a new Client

We will now load the previously saved Client snapshot into a new one. We will do this by calling load_from_json_file

new_client = Client.load_from_json_file(filepath = "ax_client_snapshot.json")

2. Validate the state of the attached experiment via summarize()

We can now view the state of the experiment by calling summarize(), and validate that it is the same as the one we saved in the earlier section

new_client.summarize()

Customizing the Serialization Process

If you have custom metrics or runners that you want to ensure are saved to JSON properly, you can initialize the Client with a StorageConfig that contains a RegistryBundle, that bundles together encoding and decoding logic for use in the save/load functions.

storage_config = StorageConfig(
    registry_bundle = RegistryBundle(
        runner_clss={MyRunner: None},
        metric_clss={MyMetric: None},
    )
)

client = Client(storage_config = storage_config)

Learn more

Take a look at these other recipes to continue your learning:

• Saving, Loading, and Updating an Ax Experiment from SQLite