[Bug]: RangeParameterConfig error when using some float stepsize

[AlbanMor]

What happened?

Hi,
First of all, Thank you for this amazing package!

Here is the issue. When trying to use a step size of 0.1 in the range of [8 to 9.8] in a RangeParameterConfig, when trying to configure the experiment, it throws the error "The range of the parameter must be evenly divisible by the step size." when trying to configure the experiment.

The issue is that in that case for some reason (upper - lower) % step_size != 0. It is equal to 6.106226635438361e-16...

Please provide a minimal, reproducible example of the unexpected behavior.

from ax.api.configs import RangeParameterConfig

test = RangeParameterConfig(name="test", parameter_type="float", bounds=(8, 9.8), step_size=0.1)

print((9.8 - 8) % 0.1)

client = Client()
client.configure_experiment(parameters=[test])

Please paste any relevant traceback/logs produced by the example provided.

Ax Version

1.2.1

Python Version

3.13.3

Operating System

MS

(Optional) Describe any potential fixes you've considered to the issue outlined above.

Might be worth changing the code for if (upper - lower) % step_size > 1e-14. But I don<t know what would be the implications further down the road in your package.

In the mean time I'll just be changing my units ;-).

Pull Request

None

Code of Conduct

• I agree to follow Ax's Code of Conduct

[mgrange1998]

Hi @AlbanMor, thank you for using Ax and for opening this issue!

This is a known bug (#4599) and a PR was landed to fix this issue #4604. I reproduced w/ your code on the latest build to ensure it would be fixed.

This change was not picked up in the latest Ax Version 1.2.1, but the next Ax release will include this change.
For the time being, thank you for your flexibility in changing to work around this issue!