New release on PyPi?

[maximlt]

The last release on PyPi - pydensecrf 1.0rc3 - dates from 2018. Is there a new release planned?

I believe this would help fixing the automatic builds attempted by conda-forge (https://github.com/conda-forge/pydensecrf-feedstock/pulls). conda is a great way to install a library like pydensecrf whose install process can be challenging for some users given their skills and the platform they use.

[lucasb-eyer]

Thanks for the ping, if you think that'd be useful, I can try to find some time to make one this week-end!

[maximlt]

I think that would be great! Don't hesitate to ping me whenever this happens, I could then look into updating the recipe on conda-forge and see how that goes.

[maximlt]

Just out of curiosity, is there any news on this topic?

[rsignell-usgs]

@lucasb-eyer, just another shout out that a new release on pypi would be very helpful! Thanks for the consideration!

[maximlt]

I think that a solid first step here would be to define what the next version is going to be called, maybe1.0, and then update the setup.py file accordingly. Tagging that commit would be helpful to, but not required.

One oddity, there are three releases on PyPi: 1.0rc1, 1.0rc2 and 1.0rc3, while on the repo the version declared in the setup.py file is 1.0rc2, so not the latest:

pydensecrf/setup.py

Line 23 in 0d53acb

version="1.0rc2",

I'm suggesting this step as I believe that I could actually get a conda-forge recipe in place pointing at the repo directly instead of at PyPi (having a release on PyPi would be the best approach of course).

[ocefpaf]

I'm suggesting this step as I believe that I could actually get a conda-forge recipe in place pointing at the repo directly instead of at PyPi (having a release on PyPi would be the best approach of course).

Hi there, just a few comments from the conda-forge side:

1. we can point to a tag/release here in the repo, no need to use PyPI (although that is recommend to increase the reach of your package to all Python users out there);
2. It is relatively easy to automate the PyPI auto-publish when tagging a new release here in the repo, reducing the load maintenance;
3. we can also tie the package version to the tag to avoid the version mismatch, it will be defined in one place and propagated to the right places where that metadata is required.

I can help with 1 when #116 is merged. I can also send PRs to implement 2 and 3 if you are interested in automating those parts.

[rsignell-usgs]

Ping @maximlt just to get this back on the radar! :)
Seems not much is left to do, right?

[rsignell-usgs]

Oops, sorry @maximlt ! It's @lucasb-eyer who can save the day!

[ddbuscombe]

@lucasb-eyer apologies for the ping, but it would greatly help our cause if you are able to merge #116 so @ocefpaf can make a new conda-forge release

We are building a whole family of end-to-end data-model-pipeline tools for to carry image segmentation (user-interactive and automated) including this and this and even this for mapping with underwater sonar, that are quite reliant on your excellent pydensecrf port. While these repos don't have many stars (because they are used by academics mostly), they are in fact used by many research groups and all of these tools are in danger of becoming hard to manage because of the current python 3.6 limitation ...

I also hope to revive some professional teaching materials that have used pydensecrf for 4+ years (I could go on - I have been a heavy user!)

This will benefit a lot of users, other than us - pydensecrf seems to be the go-to library for a great number of applications involving densification of spare labels conditional on image features, and post-processing of raw softmax scores from ML segmentation models

Thanks for your consideration!

[ddbuscombe]

This request is now 161 days old @lucasb-eyer

[ddbuscombe]

We would be very grateful for your attention

[rsignell-usgs]

@dbuscombe-usgs, @maximlt and @ocefpaf , perhaps it would be best to fork this project so we can get a new release on conda-forge?