Closed drbenvincent closed 5 years ago
I think this is mostly resolved now. We can get design spaces appropriate for the magnitude effect. It also runs being called from PsychoPy. The only niggle is that it's not utilising the full range of DB
values. I think this might be a case of design selection, and the still to be completed completion of the heuristics (see #27)
Yes, the point of this issue (that there was a bug) is now resolved. Some other issue will ensure we have sensible heuristics to guide the design optimisation (#27)
So far we do not have multiple
RB
values when choosing magnitude effect modelsIn fact, there is a bug which will crash things when we enter multiple
RB
values. Need to fix this inDARC_Designs.__innit__()
chooser.py
test_designs.py
DARC_Designs.__innit__()
.chooser.py
This relates to #21 in that we need robust validation and checking of desired design space specs etc.
__innit__()
and have a separate validator?Update
This is happening because this Python implementation encodes the design space directly (as in you ask for combinations of RA, DA, PA, RB, DB, PB), but in the Matlab implementation we had RA_over_RB instead of specifying RA directly. This new approach is a move forward, but the RA_over_RB does make sense for magnitude effect models. Why? Because if we have RB values of 10, 100, 1000 etc, it makes most sense for RA to be uniformly spaced as a proportion RB. Small increments make sense when RB is 10, but makes no sense when RB is 10000.
What to do? Allow class to accept an optional RA_over_RB list.
DARCDesign.__init__
to deal withRA_over_RB
being provided as an argument.RA_over_RB
in how we generate the initial set of designs.