Question: how to improve guessing (research software or not)?

[pawamoy]

Here are simple things I thought of to improve the guessing as research software, please tell us if they actually impact the guessing or not (and add the ones you know/thought of as well :smile:):

• Use the Intended Audience :: Science/Research PyPI classifier in setup.py.
• Add research, science, university or other keywords in README.
• Add those keywords in the keywords section of setup.py.
• More generally, add those keywords anywhere in the documentation (including source code doc like docstrings).
• Again, add those keywords in the tags (topics) of the GitHub repository.
• Add citations, links to papers (articles) or bibliography in the documentation.
• Have well-known university email addresses or university websites links.

Note: I couldn't find any docs on how Depsy is guessing that, but maybe I missed something?

[andim]

To give a concrete example depsy does not classify Noisyopt (https://github.com/andim/noisyopt) as research software http://depsy.org/package/python/noisyopt

And this despite Intended Audience :: Science/Research tag on pypi and clearly scientific aim.

If anyone has an idea of what is going, let me know!

[pawamoy]

Alright, I'm taking a look at the source code.

Intended Audience :: Science/Research classifier should work. If it's not the case (and we see that with your Noisyopt package), I suspect this line

if package_obj.intended_audience == "Science/Research":

to return False because of a trailing new-line (maybe), or the intended_audience property to return None:

try:
    pypi_classifiers = self.api_raw["info"]["classifiers"]
except KeyError:
    return None

I also see in the code that a list of 'sciency' keywords are used to tag a package as academic, see the list here: https://github.com/Impactstory/depsy/blob/master/models/academic.py#L29.

However, these keywords are tested against a truncation of the 100 first characters of the package description:

self.summary = truncate(self.api_raw["info"]["description"])

They also are tested against the package's name and tags obtained through setup's keywords and classifiers information.

So here is the updated list of criteria:

• [x] Intended Audience :: Science/Research classifier: YES, should work but seems to be buggy.
• [x] Keywords in short description: YES, if keywords appear in the first 100 characters.
• [x] Keywords in keywords section of setup.py: YES.
• [x] Keywords in project name: YES.
• [ ] Keywords in README: NO.
• [ ] Keywords in long description: NO.
• [ ] Keywords in source code or documentation: NO.
• [ ] Keywords in GitHub repository's topics: NO.
• [ ] Citations / bibliography in the documentation: NO.
• [ ] University email addresses / links: NO.

The unchecked criteria could serve as a TODO-list :smile:

[pawamoy]

Closing because project is not maintained anymore.