zero_method 'wilcox' and 'pratt' do not work if the x - y is zero for all elements.

[DennisShaw]

ValueError: zero_method 'wilcox' and 'pratt' do not work if the x - y is zero for all elements.

[naman-ali]

@DennisShaw Were you able to solve this issue?

[GTFoster]

I've also run into the same issue. I got this error when trying to run the first bash command suggested by the readme file after cloning the repository.

[hajarmerbouh]

The error means that either wilcoxon or pratt is having as an input two identical lists : For a given list=List1, wilcoxon(List1,List1) returns error; in other words, x-y is zero for all elements which points out the erreur.

In order to avoid it, we'll have to make sure that repo[i]!=repo[j] (the inputs of wilcox). So we need to apply the following adjustements (mentionned in bold):

[267] for i in range(type_size-1):
print ("repo ",i, ": ",repo[i],type(repo[i]))
for j in range(i+1, type_size):
if evaluation_metric == 1:
etc..
This way, 'i' will always be different from 'j' and so repo[i] and repo[j] won't be identical.

[abbynewbury]

I just wanted to check in regarding this issue, if the above comment is the verified solution to the issue and should be implemented? Or was there another method that the authors intended and that is why this comment status is left on 'Open'. Thanks!

[BinglanLi]

The issue arises from how statistical tests return test results for identical inputs. The original code seems to assume that the statistical tests will return nan and subsequently capture it usingif result != result.

Here is my proposed fix:

#pairwised judgement
def wilcoxon_test(v1,v2):# original metric: the smaller the more similar
    if v1 == v2:
        return 1/(math.sqrt(0)+1)

    result = stats.wilcoxon(v1, v2).statistic
    if result != result: # this originally checks for nan result from the stat test, but now many test will raise an error warning, rather than return nan
        result = 0
    return 1/(math.sqrt(result)+1)

def entroy_test(v1,v2):#original metric: the smaller the more similar
    if v1 == v2:
        return 0

    result = stats.entropy(v1,v2)
    result = stats.wilcoxon(v1, v2).statistic
    if result != result:
        result = 0
    return result

def spearmanr_test(v1,v2):#original metric: the larger the more similar
    if v1 == v2:
        return sigmoid(-1)

    result = stats.mstats.spearmanr(v1,v2).correlation
    # print(f'v1 and v2 correlation: {result}')
    # print(f'v1 = {v1}, v2 = {v2}')
    result = stats.wilcoxon(v1, v2).statistic
    if result != result:
        result = -1
    return sigmoid(result)

def pearsonr_test(v1,v2):#original metric: the larger the more similar
    if v1 == v2:
        return sigmoid(-1)

    result = stats.mstats.pearsonr(v1,v2)[0]
    result = stats.wilcoxon(v1, v2).statistic
    if result != result:
        result = -1
    return sigmoid(result)

I don't think if result != result is necessary anymore, but I still left it here to capture other cases I've not thought through.