mstats.winsorize documentation needs clarification

[pv]

stats.mstats.winsorize documentation needs some clarification:

From #7717 (comment):

This shows the difference (look at the number of values at start/end that are identical):

In [11]: data
Out[11]: 
masked_array(data = [77 87 88 114 151 -- 219 246 253 262 296 299 306 376 428 515 666 1310 2611],
             mask = [False False False False False  True False False False False False False
 False False False False False False False],
       fill_value = 999999)

In [12]: mstats.winsorize(data, (0.2, 0.2))
Out[12]: 
masked_array(data = [114 114 114 114 151 515 219 246 253 262 296 299 306 376 428 515 515 515
 515],
             mask = [False False False False False False False False False False False False
 False False False False False False False],
       fill_value = 999999)

In [13]: mstats.winsorize(data, (0.2, 0.2), inclusive=(False, False))
Out[13]: 
masked_array(data = [151 151 151 151 151 428 219 246 253 262 296 299 306 376 428 428 428 428
 428],
             mask = [False False False False False False False False False False False False
 False False False False False False False],
       fill_value = 999999)

I think it's the description of inclusive in the docstring that's a little confusing, it talks about number of data being masked while the truncation is about indices used to construct the lowidx-upidx interval of data to keep. Can you try to reformulate that description?

[deepakgouda]

According to the documentation, if inclusive is True then the number of data being masked on each side should be rounded and if False, then it should be truncated. But the code in scipy/scipystats/mstats_basic.py line 1763

if low_include:          #low_include is the first entry of the inclusive tuple
       lowidx = int(low_limit * n)
else:
       lowidx = np.round(low_limit * n).astype(int)

does the opposite.

So, either the documentation has to be changed or the if clause. Shall I make a PR?