Open AllenDowney opened 6 years ago
In general the best approach is to bump the conda-forge feedstock. The CI there helps iron out any recipe problems, and we can pull in changes very directly now. You can also submit PRs to the recipes at AnacondaRecipes: https://github.com/AnacondaRecipes/statsmodels-feedstock, but submitting it to conda-forge is preferable, thanks to their CI setup. We will see to this update.
Mike, thanks for the quick response. I don't understand most of what you just said. Is this something that will be fixed soon, or are you telling me how to work around it?
On Fri, Mar 23, 2018 at 10:46 AM, Mike Sarahan notifications@github.com wrote:
In general the best approach is to bump the conda-forge feedstock. The CI there helps iron out any recipe problems, and we can pull in changes very directly now. You can also submit PRs to the recipes at AnacondaRecipes: https://github.com/AnacondaRecipes/statsmodels-feedstock, but submitting it to conda-forge is preferable, thanks to their CI setup. We will see to this update.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/ContinuumIO/anaconda-issues/issues/8979#issuecomment-375688437, or mute the thread https://github.com/notifications/unsubscribe-auth/ABy37RAM-Sjvd5X8aWvQfG4QBpGHvFuwks5thQqsgaJpZM4S3HIr .
Sorry, just looked at this, and it seems like statsmodels 0.9 is not released yet. We can't update packages until then. We can hotfix the index metadata to introduce a constraint so that statsmodels 0.8 depends on scipy <1.0.
The Anaconda packages are produced by recipes. There used to be a private internal pool of these. Now they're all public, and they are all drawn from conda-forge, which is a community organization for building packages. They created some clever automation that takes a lot of the burden out of the trial and error of package building. What I pointed you to is two places where you can edit the recipe (to do the update). With conda-forge, you'd submit a PR to their https://github.com/conda-forge/statsmodels-feedstock repo, and once that PR is merged, your new packages would appear on the conda-forge channel automatically.
With those packages proven, we can pull in the recipe changes to Anaconda, and build them on our build system, which we keep separate for the sake of security guarantees for our customers.
Thanks for the explanation!
On Fri, Mar 23, 2018 at 10:54 AM, Mike Sarahan notifications@github.com wrote:
The Anaconda packages are produced by recipes. There used to be a private internal pool of these. Now they're all public, and they are all drawn from conda-forge, which is a community organization for building packages. They created some clever automation that takes a lot of the burden out of the trial and error of package building. What I pointed you to is two places where you can edit the recipe (to do the update). With conda-forge, you'd submit a PR to their https://github.com/conda-forge/statsmodels-feedstock repo, and once that PR is merged, your new packages would appear on the conda-forge channel automatically.
With those packages proven, we can pull in the recipe changes to Anaconda, and build them on our build system, which we keep separate for the sake of security guarantees for our customers.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/ContinuumIO/anaconda-issues/issues/8979#issuecomment-375691102, or mute the thread https://github.com/notifications/unsubscribe-auth/ABy37XXyavsCJ1Cq6dApYIKsqzTMCyfCks5thQyfgaJpZM4S3HIr .
It looks like the current versions of scipy and statsmodels are
scipy 1.0.0 py36hbf646e7_0
statsmodels 0.8.0 py36h8533d0b_0
With this combination, we have the problem described here:
https://github.com/statsmodels/statsmodels/issues/3931
statsmodels 0.8 uses chisqprob, which is no longer in scipy.
Apparently statsmodels 0.9 has resolved this issue, so this problem should be resolved by the next update of statsmodels. Any idea when that will be?
Thanks!
Anaconda or Miniconda version:
Operating System: Linux
conda info
conda list --show-channel-urls