|
| 1 | +# Specifying Outcome Constraints |
| 2 | +## Introduction |
| 3 | +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. |
| 4 | + |
| 5 | +## Prerequisites |
| 6 | +Before specifying outcome constraints, make sure you have an understanding of Ax [experiments](#) and their components. |
| 7 | + |
| 8 | + |
| 9 | +We will also assume you are already familiar with |
| 10 | +[using Ax for ask-tell optimization](#), though this can be used for closed-loop |
| 11 | +experiments as well. |
| 12 | + |
| 13 | + |
| 14 | +## Setup |
| 15 | +Before we begin you must instantiate the `AxClient` and configure it with your |
| 16 | +experiment. |
| 17 | + |
| 18 | + |
| 19 | +```python |
| 20 | +from ax.service.ax_client import AxClient |
| 21 | + |
| 22 | +client = AxClient() |
| 23 | +``` |
| 24 | + |
| 25 | +## Steps |
| 26 | + |
| 27 | +1. Define the Metrics You Want to Constrain |
| 28 | +2. Create Outcome Constraint Map |
| 29 | +3. Add the Constraints to the Experiment |
| 30 | + |
| 31 | + |
| 32 | +### 1. Define the Metrics You Want to Constrain |
| 33 | +Construct a list of `metric_names` to specify the metrics you intend to constrain. |
| 34 | + |
| 35 | +```python |
| 36 | +# Define the metrics to constrain |
| 37 | + |
| 38 | +metric_names=["metric_1", "metric_2"], |
| 39 | +``` |
| 40 | + |
| 41 | +### 2. Create Outcome Constraint Map |
| 42 | +Create an `outcome_constraints` dictionary with the constraint information. |
| 43 | + |
| 44 | +```python |
| 45 | +outcome_constraints = ["metric_1 <= -1%"] # the % makes it a relative constraint |
| 46 | +``` |
| 47 | + |
| 48 | +Alternatively, you can create multiple outcome constraints: |
| 49 | + |
| 50 | + |
| 51 | +```python |
| 52 | +outcome_constraints = [ |
| 53 | + "metric_1 <= -1%", |
| 54 | + "metric_2 >= 0.5%" |
| 55 | +] |
| 56 | +``` |
| 57 | + |
| 58 | +### 3. Add the Constraints to the Experiment |
| 59 | +Call the `set_optimization_config` method, passing in the list of constraints. |
| 60 | + |
| 61 | + |
| 62 | +```python |
| 63 | +from ax.core.objective import ObjectiveProperties |
| 64 | + |
| 65 | +objectives = { |
| 66 | + 'metric_1': ObjectiveProperties(minimize=True) |
| 67 | +} |
| 68 | +client.set_optimization_config( |
| 69 | + objectives=objectives, |
| 70 | + outcome_constraints=outcome_constraints, |
| 71 | +) |
| 72 | +``` |
| 73 | + |
| 74 | +### Learn More |
| 75 | +For further learning, explore these additional resources: |
| 76 | + |
| 77 | +* [Creating tracking metrics in Ax](#) |
| 78 | +* [Creating optimization configurations in Ax](#) |
0 commit comments