search

giotto-ai / giotto-tda

A high-performance topological machine learning toolbox in Python
https://giotto-ai.github.io/gtda-docs
Other
848 stars 173 forks source link

Multiple names on pypi #311

Closed rth closed 2 years ago

rth commented 4 years ago

I think the current situation where both giotto-tda, giotto-lean, giotto-learn-nightly and giotto-tda-nightly can be confusing.

For instance, I just searched for "giotto nightly" and ended up installing the wrong version of the nightly: giotto-learn-nightly

While maintaining previous names for backward compatibility can make sense, at least for some time, I think it would be helpful to push a package generated with https://pypi.org/project/pypi-alias/ to giotto-learn* packages, so that users using that name still end up with the latest version of giotto-tda, and the readme on PyPi should explicitly indicate that is is the wrong name.

In the long term this does pose a searching issue, as for instance searching for "giotto pypi" does bring up the wrong version first.

rth commented 4 years ago

Also there should be no harm, in at least removing giotto-learn-nightly I think.

ulupo commented 4 years ago

Thanks @rth! Concerning giotto-learn-nightly, we are keeping it for legacy reasons due to what arguagly has been a development mistake in introducing a feature too early. This feature was widely advertised (as part of a nightly release) in https://towardsdatascience.com/the-shape-of-chemical-functions-d1e1568d020, but it has yet no place in giotto-tda (even nightly).

Concerning having giotto-learn, I just want to point out that the latest version pushed will warn the user that they are dealing with a defunct project, upon any import. See https://github.com/giotto-ai/giotto-tda/tree/giotto-learn_0.1.3-r1. This is how we currently deal with users landing on the wrong PyPI package.

rth commented 4 years ago

the user that they are dealing with a defunct project, upon any import.

Indeed, but the user will import gtda which doesn't exist.

This feature was widely advertised (as part of a nightly release) in towardsdatascience.com/the-shape-of-chemical-functions-d1e1568d020,

Aww, I see.

ulupo commented 4 years ago

Indeed, but the user will import gtda which doesn't exist.

Fair point. The rationale was that at this point we have essentially erased all traces of giotto-learn aside from that very Medium blog post which I referred to. So we assumed virtually nobody would want to run pip install giotto-learn. But your Google search example seems to disprove that claim. We'll discuss this.