Add outcome constraints recipe (facebook#3530)

Summary:

as titled - any edits are greatly appreciated!

Differential Revision: D71403614

Author: paschai

docs/recipes/outcome-constraints.md

@@ -0,0 +1,55 @@
+# Specifying Outcome Constraints
+## Introduction
+Outcome constraints can be a crucial component of optimization in Ax. They allow you to specify constraints on the outcomes of your experiment, ensuring that the optimized parameters do not degrade certain metrics.
+
+## Prerequisites
+Before we begin you must instantiate the `Client` and configure it with your experiment and metrics.
+
+We will also assume you are already familiar with [using Ax for ask-tell optimization](#).
+
+```python
+client = Client()
+
+client.configure_experiment(...)
+client.configure_metrics(...)
+```
+
+## Steps
+
+1. Configure an optimization with outcome constraints
+2. Continue with iteratinf over trials and evaluating them
+
+### 1. Configure an optimization with outcome constraints
+We can leverage the Client's `set_optimization_config` method to configure an optimization with outcome constraints. This method takes an objective string and an outcome constraints string.
+
+Outcome constraints allow us to express a desire to have a metric clear a threshold but not be further optimized. These constraints are expressed as inequalities.
+
+```python
+client.configure_optimization(objective="testObjective", outcome_constraints=["qps >= 100"])
+```
+
+To indicate a relative constraint, multiply your bound by "baseline":
+```python
+client.configure_optimization(
+    objective="testObjective",
+    outcome_constraints="qps >= 0.95 * baseline"
+)
+```
+This example will constrain the outcomes such that the QPS is at least 95% of the baseline arm's QPS.
+
+Note that scalarized outcome constraints cannot be relative.
+
+### 2. Continue with iterating over trials and evaluating them
+Now that your experiment has been configured for a multi-objective optimization, you can simply continue with iterating over trials and evaluating them as you typically would.
+
+```python
+trial_idx, parameters = client.get_next_trials().popitem()
+client.complete_trial(...)
+```
+
+## Learn more
+
+Take a look at these other resources to continue your learning:
+
+- [Multi-objective Optimizations in Ax](#)
+- [Scalarized Objective Optimizations with Ax](#)

website/sidebars.js

@@ -58,6 +58,7 @@ export default {
       'recipes/experiment-to-sqlite',
       'recipes/multi-objective-optimization',
       'recipes/scalarized-objective',
+      'recipes/outcome-constraints',
     ],
   },
 };