Skip to content

Deployments

Models are automatically deployed if their key metric (R2 for regression, F1 for classification) is improved over the currently deployed version during training. If you want to manage deploys manually, you can always change which model is currently responsible for making predictions.

API

pgml.deploy
1
2
3
4
5
pgml.deploy(
    project_name TEXT,                  -- Human-friendly project name
    strategy TEXT DEFAULT 'best_score', -- 'rollback', 'best_score', or 'most_recent'
    algorithm_name TEXT DEFAULT NULL    -- filter candidates to a particular algorithm, NULL = all qualify
)

The default behavior allows any algorithm to qualify, but deployment candidates can be further restricted to a specific algorithm by passing the algorithm_name.

Note

Deployed models are cached at the session level to improve prediction times. Active sessions will not see deploys until they reconnect. 

1
SELECT * FROM pgml.deploy('Handwritten Digit Image Classifier', 'best_score');

1
2
3
4
            project_name            |    strategy    | algorithm_name
------------------------------------+----------------+----------------
 Handwritten Digit Image Classifier | classification | linear
(1 row)

Strategies

There are 3 different deployment strategies available

strategy description
most_recent The most recently trained model for this project
best_score The model that achieved the best key metric score
rollback The model that was previously deployed for this project

Rolling back to a specific algorithm

Rolling back creates a new deploy for the model that was deployed before the current one. Multiple rollbacks in a row will effectively oscilate between the two most recently deployed models, making rollbacks a relatively safe operation. 

1
SELECT * FROM pgml.deploy('Handwritten Digit Image Classifier', 'rollback', 'svm');

1
2
3
4
            project_name            |    strategy    | algorithm_name
------------------------------------+----------------+----------------
 Handwritten Digit Image Classifier | classification | linear
(1 row)

Comments