Closed ihincks closed 6 years ago
I'm done with what I wanted to do in this PR (aside from doc build error fix). Thanks to @whitewhim2718 and his feature-generalized-outcomes branch from which I copied some ideas and code.
As usual, I find the doc error messages pretty cryptic, and don't know where to start at the moment.
I took a look at your error log. All the red lines seem to be warnings. I'm pretty sure the error failing the build is make[1]: latexmk: Command not found
on line 1564 (https://travis-ci.org/QInfer/python-qinfer/jobs/242525398#L1564).
The last master build does this:
Running LaTeX files through pdflatex...
$ make -C _build/latex all-pdf
make[1]: Entering directory `/home/travis/build/QInfer/python-qinfer/doc/_build/latex'
$ pdflatex 'QInfer.tex'
And your build is doing this:
Running LaTeX files through pdflatex...
$ make -C _build/latex all-pdf
make[1]: Entering directory `/home/travis/build/QInfer/python-qinfer/doc/_build/latex'
$ latexmk -pdf -dvi- -ps- 'QInfer.tex'
Taking a look at the installed dependencies, I see master installed sphinx==1.5.6
and your build installed sphinx==1.6.2
, which might be responsible.
Thanks for looking into this, @scasagrande . It seems that they switched to latexmk recently: https://github.com/sphinx-doc/sphinx/issues/3543
Bah, need to somehow turn off interactive mode in latexmk, the build has stalled. Either that, or revert to an older version of sphinx, which is probably not a great idea.
I'd lock it to the older version for now, and then make it a separate PR to update to the latest.
Looking at the Travis build failures, it seems as though the latex_preamble
Sphinx configuration key is not being propagated correctly in the newest version. It appears that the key is now latex_elements["preamble"]
as of 1.5.1, so it might work to provide the preamble in both forms. I'll try that and PR it now.
This should now work with #135 merged in. Resolved conflicts and re-ran tests.
It would be nice for
bayes_risk
to vectorize over inputs to make use of any speed-ups that the likelihood might have along itsexpparams
axis. This PR is a stab at that. It seems to be working for the model I am working on, but needs a bit more testing.