[Catalog-sig] Please turn off ratings

Jacob Kaplan-Moss jacob at jacobian.org
Wed Apr 6 01:48:43 CEST 2011


[Oops, forgot to reply all - sorry for the duplicate, Martin!]


On Tue, Apr 5, 2011 at 4:42 PM, "Martin v. Löwis" <martin at v.loewis.de> wrote:
> To put some objective data into the discussion: here is the number of
> ratings that got posted recently, by month:
>
>  2010 |  9 |    41
>  2010 | 10 |    51
>  2010 | 11 |    42
>  2010 | 12 |    51
>  2011 |  1 |    76
>  2011 |  2 |    71
>  2011 |  3 |    47

Thank you.

Here are some more numbers: I took a look [1] at how many ratings
packages have received. Here's the distribution (# of ratings: # of
packages):

0: 13452
1: 427
2: 91
3: 22
4: 12
5: 10
6: 3
7: 8
8: 4
9: 2
10: 1
11: 2
13: 1
21: 1

So 96% of packages have no ratings. Only 1% have more than one rating.

I ask again, how is this useful?

Django, with 7 ratings, is in the 99.9th percentile among all
packages, so if anyone could find this useful I suppose it might be
me. But 7 ratings isn't in any way a significant sample: PyPI shows
11,000 downloads of Django 1.3 alone (which is just a few weeks old);
there's no *way* this provides *any* useful data whatsoever.

Jacob

[1] Using this script::

from collections import defaultdict
import xmlrpclib

pypi = xmlrpclib.ServerProxy('http://pypi.python.org/pypi')
all_packages = pypi.list_packages()

all_counts = defaultdict(int)
for p in all_packages:
ratings = pypi.ratings(p, "", 0)
print p, len(ratings[0])
all_counts[len(ratings[0])] += 1

open('counts.txt', 'w').write(str(all_counts))
print all_counts


More information about the Catalog-SIG mailing list