Code to select random conditions is incorrect

[carolfs]

From PsychoPy's documentation:

The parameter Select rows allows this. You can specify which rows you want to use by inserting values here:

• 0,2,5 gives the 1st, 3rd and 5th entry of a list - Python starts with index zero)
• random(4)*10 gives 4 indices from 0 to 10 (so selects 4 out of 11 conditions)

The first item is incorrect: 0,2,5 gives the 1st, 3rd, and 6th entries.

The second item is just plain wrong as far as I understand. numpy's random function generates random numbers in the interval [0, 1), thus random(4)*10 (which by the way should be $random(4)*10) generates 4 floating-point numbers in that interval. Multiplying by 10 thus generates 4 floating-point numbers in the [0, 10) interval. So these numbers are floats, not integers. But as far as I understand PsychoPy converts them to integers like this:

    # the selection might now be a slice or a series of indices
    if isinstance(selection, slice):
        trialList = trialList[selection]
    elif len(selection) > 0:
        allConds = trialList
        trialList = []
        for ii in selection:
            trialList.append(allConds[int(round(ii))])

Thus the numbers are rounded before being converted to integers. This means that numbers from 0 to 0.5 are converted to 0, numbers from 0.5 to 1.5 are converted to 1 and so on. Thus, there's a lower probability that 0 or 10 will be selected.

The probability for every number should be 1/11, but it's not.

In my opinion, int(round(ii)) should be changed to int(np.floor(ii)) and the documentation to "random(4)*10 gives 4 indices from 0 to 9 (so selects 4 out of 10 conditions)." This will make sure every number has the same probability of being selected.

[peircej]

I agree on all the above points, which were clearly expressed 👏 I think those docs have been written in a slight rush (on a relatively rarely-used feature so haven't been noticed) and aren't quite right.

Are you comfortable submitting a pull request with the changes? (The docs source is in the repo as well)

[carolfs]

I will try to submit one later today.