[GENERAL SUPPORT]: Question regarding RangeParameters.digits

[delfoxav]

Question

Hello,

For my specific setup, I am using AX with continuous chemical concentrations. Considering the physical limitation such as pipetting accuracy, we cannot consider really low concentrations. Thus it would be interesting to "round" the trials sampled parameters to a certain number of digits.

So far, we have been applying a rounding on the returned values of AX. However that approach means that there's a small difference between what AX expects and what we are really testing.

As a simple example, let's say we want to test some NaCl concentration within the range (0, 50) and that we have a precision of 0.1 mM. Ax could propose something such as 30.2500000001 and 30.34999999999, these two values are different for AX, but for our physical process they both results in 30.3 mM. What are the risk of that difference?

Futhermore, while exploring the source code, I saw the RangeParameter class used by RangeParameterConfig has a "digits" attributes that could be set and would give similar results. Is there a reason why "digits" isn't present in RangeParameterConfig ? What approach would you recommend?

Thanks for the help,

Best,

Xavier

Please provide any relevant code snippet if applicable.

Code of Conduct

• I agree to follow this Ax's Code of Conduct

[hvarfner]

Hi @delfoxav ,

Sorry for the late reply! In your specific case when the difference is that small (i.e. 0.2% of one parameter's range), you are probably going to be fine. If you want a hard and fast rule, If you have 2+ parameters and the rounding is less than 1%, I would personally say that it is not a major concern that you don't account for the rounding.

Re. RangeParameter vs. RangeParameterConfig: You have a step_size argument for RangeParameterConfig, which IIUC is exactly what you would want and what I would recommend.