Bug in Resolutions used for Fractional Factorial Design

[avanelc]

I encountered an issue with the add_design method in the doepandas package when specifying resolutions for fractional factorial designs with 6 factors. The method allows for the selection of a resolution that is not standard, resulting in an error.

Issue Details:
When attempting to create a design with resolution=VI for 6 factors, the method fails with a ValueError indicating "design not possible." But when setting resolution=V for 6 factors, the design given is with 32 runs. Although, resolution V does not typically exist for 6 factors in standard design tables.

Standard Resolutions:
For 6 factors, the standard resolutions are:
Resolution IV: 16 runs (1/4 fraction of the full factorial)
Resolution VI: 32 runs (1/2 fraction of the full factorial)

Expected Behavior:
The method should either generate a valid design for compatible resolutions or provide a descriptive error message indicating that the requested resolution is not valid for the given number of factors.

[relf]

Hi, could you provide a simple code to reproduce the issue? By the way, I did not find anything about a doepandas package on Pypi.

[avanelc]

Hi, could you provide a simple code to reproduce the issue? By the way, I did not find anything about a doepandas package on Pypi.

Hello! Sorry I was looking at another method, I meant to write 'fracfact_by_res' method from pyDOE3

Sure! Let's consider an example from a manufacturing context with measurable factors. Here are six factors, each with two levels, expressed numerically and with units:

Factors and Levels:
Machine Speed: 100 RPM (Level 1), 200 RPM (Level 2)
Temperature: 150°C (Level 1), 200°C (Level 2)
Pressure: 10 Bar (Level 1), 20 Bar (Level 2)
Material Thickness: 1 mm (Level 1), 2 mm (Level 2)
Cooling Rate: 5°C/min (Level 1), 10°C/min (Level 2)
Feed Rate: 1 m/min (Level 1), 2 m/min (Level 2)

[relf]

I should have mentioned that I do not know the function fractfact_by_res() (and I do not know anything about the standard table you mentioned). For the record, I've forked pyDOE mainly out of my need for LHS and Fullfactorial DOEs.

From what you stated, I understand that fractfact_by_res(n=6, res=5) should raise ValueError instead of returning 32-points doe while fractfact_by_res(n=6, res=6) should return a 32-points doe instead of ValueError. Correct?

Feel free to contribute with a PR as I think I will have no time to dig into this.

[relf]

I am wondering if this issue could be fixed by this pending PR in pyDOE2.

[avanelc]

I think that the reintegration of the bofire code on pyDoE3 (#11) would solve this issue.

[relf]

Did you test? Would you mind contributing with a PR?