Lieber Fred,

I sometimes encounter NaNs in the log(target) in corners of the parameter space. It would be very helpful, if this modification or something similar to it could end up in a released version of pypmc.

I am also happy to change the name of the keyword argument if you disagree. I had considered e.g. ignore_NaN. Let my know what you think!

Cheers Danny

Installing pypmc with pip under Python 3.7 fails, see e.g. here for an error log.

I suspect re-generating the Cython files with a more up-to-date Cython version is sufficient to fix it.

(Of course, having wheels on PyPI would be even cooler...)

Cheers! D.

Hi Fred,

This PR does two things:

It addresses part of #67, with the solution I saw at https://github.com/aresch/rencode/pull/25/files

And it fixes a type bug I reported before: https://github.com/fredRos/pypmc/pull/59

Is having the secret deploy keys in the github repository intentional?

Cheers, Johannes

As @dvandyk realized in the EOS code, there is a global misunderstanding of whether PMC updates should be performed on the last step samples/weights or on all the samples/weights. The misunderstanding is also present in the documentation: https://github.com/pypmc/pypmc/blob/5f705e9c67add23e46999f8b3c2483e1593f3683/examples/pmc.py#L67-L69

By changing [:] by [-1] as it should be from the comment, the efficiency of the adaptation is greatly improved! A perplexity of 0.99 is obtained after 5 steps vs. 0.53 after 10 steps in the present version of the example!

• [x] update links in docs to new URL pypmc.github.io/pypmc
• [x] update theme: old matplotlib theme looks dated now
• [x] update copyright notice: definitely year, authors?
• [ ] ~~have a fresh look: text out of date?~~
• [ ] ~~add section with testimonials or others using pypmc, either in docs or in readme~~

Whenever I run the setup.py through pip or directly through python it gives me this error: I went into the setup.py and fiddled with the code and found that these lines were causing the errors: It allows me to install pypmc if I delete the commands inside the extra_compile_args function, but then it doesn't function properly as it can not import the name 'gauss'.

Hi all,

This is just to let you know that there is now a pypmc package on conda-forge.

https://github.com/conda-forge/pypmc-feedstock https://anaconda.org/conda-forge/pypmc

I used this to build the snowline Bayesian inference package on top of it. https://johannesbuchner.github.io/snowline/

Cheers, Johannes

This pull request changes the way printing status messages is handled.

1. All print functionality is replaced by the logging standard library. Messages that were wrapped with if(verbose) are logged as information, or as warnings otherwise.
2. The new function util.log_to_stdout sets up a basic logging handler for easy standalone usage of pypmc. This function is called in __init__.py as a drop-in replacement of the current behaviour.

Comments

• Every submodule uses its own logger such that the level of granularity is still high. E.g. to only log warnings and errors from PMC, use logging.getLogger('pypmc.mix_adapt.pmc').setLevel(logging.WARNING).
• Note that the default behaviour of logging when no log handler is specified, is to print warnings and errors to stderr.
• Importing pypmc is setting up a log handler (to stdout), which must be removed explicity to use the module as a library.

WIP

• [x] Adding the default StreamHandler creates uncaptured output in the nosetests, as warnings log to stderr. To comply with the previous code, we specify to log only to stdout.
• [x] Emit deprecation notice for optional verbose argument instead of removing the option (breaking change).
• [ ] Document logging behaviour and how to use pypmc as a library. The wiki page is a good place to document logging after the proposed changes are approved.

Solves #95

Why

Python2 has reached end of life and from 1.2 and onwards, we won't test the code on that platform anymore. Occasionally, the code is more complex to support two major version of python

Acceptance criteria

• [x] All source code can be freed of compatibility constructs to make it work with python 2 and 3
• [x] the Makefile is cleaned up
• [x] the docs updated

• Uses the manylinux2014 docker image. I use the same for EOS. Can be changed later to e.g. manylinux2010 or even manylinux1?
• Builds for python3 only. Do we really need python2.x?
• Builds wheel for every commit, but only deploys to PyPI for tagged commits.
• Builds and deploys documentation for every commit. Should this step deploy only on tagged commits?

Tach @fredRos!

Since there is not tar.gz file for the v1.1.4 release on pypi, pip falls back to installing from 1.1.2 source on MacOS. I've worked around that problem for the time being my downloading the tarball and installing from the downloaded file in the EOS CI. However, it would be nice if you could make sure that 1.1.4 and future releases support from-source builds.

When a parameters is described with an angle (eg the argument of a complex number), MCMC and PMC have trouble describing the periodic boundaries of the parameter range. I see two possible solutions:

• Allow parameters to be periodic. This may require a bit of work for the correlations and correct distributions though.
• The user can provide an extended range, eg [-100*pi, 100*pi], but it would be nice if pypmc could wrap the parameter space when estimating certain quantities

Two issues in the documentation https://pypmc.github.io/user_guide.html#variational-bayes:

• The url in reference [Bis06] is broken
• \vecth and \vecgamma are not correctly displayed (at least for me)

I get the following stack trace when trying to install pypmc on Windows 10.

> pip install pypmc
Defaulting to user installation because normal site-packages is not writeable
Collecting pypmc
  Using cached pypmc-1.2.tar.gz (979 kB)
  Installing build dependencies ... done
  Getting requirements to build wheel ... done
    Preparing wheel metadata ... done
Requirement already satisfied: numpy<2.0,>=1.6 in c:\program files\python38\lib\site-packages (from pypmc) (1.19.1)
Requirement already satisfied: scipy in c:\users\astiff\appdata\roaming\python\python38\site-packages (from pypmc) (1.7.1)
Building wheels for collected packages: pypmc
  Building wheel for pypmc (PEP 517) ... error
  ERROR: Command errored out with exit status 1:
   command: 'c:\program files\python38\python.exe' 'C:\Users\astiff\AppData\Roaming\Python\Python38\site-packages\pip\_vendor\pep517\in_process\_in_process.py' build_wheel 'C:\Users\astiff\AppData\Local\Temp\tmpo5rojaiy'
       cwd: C:\Users\astiff\AppData\Local\Temp\pip-install-k386b90x\pypmc_4bb93d97d9944d4592bc69d1ad5ff28e
  Complete output (57 lines):
  running bdist_wheel
  running build
  running build_py
  creating build
  creating build\lib.win-amd64-3.8
  creating build\lib.win-amd64-3.8\pypmc
  copying pypmc\_version.py -> build\lib.win-amd64-3.8\pypmc
  copying pypmc\__init__.py -> build\lib.win-amd64-3.8\pypmc
  creating build\lib.win-amd64-3.8\pypmc\density
  copying pypmc\density\base.py -> build\lib.win-amd64-3.8\pypmc\density
  copying pypmc\density\base_test.py -> build\lib.win-amd64-3.8\pypmc\density
  copying pypmc\density\gauss_test.py -> build\lib.win-amd64-3.8\pypmc\density
  copying pypmc\density\mixture_test.py -> build\lib.win-amd64-3.8\pypmc\density
  copying pypmc\density\student_t_test.py -> build\lib.win-amd64-3.8\pypmc\density
  copying pypmc\density\__init__.py -> build\lib.win-amd64-3.8\pypmc\density
  creating build\lib.win-amd64-3.8\pypmc\mix_adapt
  copying pypmc\mix_adapt\hierarchical.py -> build\lib.win-amd64-3.8\pypmc\mix_adapt
  copying pypmc\mix_adapt\hierarchical_test.py -> build\lib.win-amd64-3.8\pypmc\mix_adapt
  copying pypmc\mix_adapt\pmc_test.py -> build\lib.win-amd64-3.8\pypmc\mix_adapt
  copying pypmc\mix_adapt\r_value.py -> build\lib.win-amd64-3.8\pypmc\mix_adapt
  copying pypmc\mix_adapt\r_value_test.py -> build\lib.win-amd64-3.8\pypmc\mix_adapt
  copying pypmc\mix_adapt\variational_test.py -> build\lib.win-amd64-3.8\pypmc\mix_adapt
  copying pypmc\mix_adapt\__init__.py -> build\lib.win-amd64-3.8\pypmc\mix_adapt
  creating build\lib.win-amd64-3.8\pypmc\sampler
  copying pypmc\sampler\importance_sampling.py -> build\lib.win-amd64-3.8\pypmc\sampler
  copying pypmc\sampler\importance_sampling_test.py -> build\lib.win-amd64-3.8\pypmc\sampler
  copying pypmc\sampler\markov_chain.py -> build\lib.win-amd64-3.8\pypmc\sampler
  copying pypmc\sampler\markov_chain_test.py -> build\lib.win-amd64-3.8\pypmc\sampler
  copying pypmc\sampler\__init__.py -> build\lib.win-amd64-3.8\pypmc\sampler
  creating build\lib.win-amd64-3.8\pypmc\tools
  copying pypmc\tools\convergence.py -> build\lib.win-amd64-3.8\pypmc\tools
  copying pypmc\tools\convergence_test.py -> build\lib.win-amd64-3.8\pypmc\tools
  copying pypmc\tools\linalg_test.py -> build\lib.win-amd64-3.8\pypmc\tools
  copying pypmc\tools\parallel_sampler.py -> build\lib.win-amd64-3.8\pypmc\tools
  copying pypmc\tools\parallel_sampler_test.py -> build\lib.win-amd64-3.8\pypmc\tools
  copying pypmc\tools\partition_test.py -> build\lib.win-amd64-3.8\pypmc\tools
  copying pypmc\tools\plot_test.py -> build\lib.win-amd64-3.8\pypmc\tools
  copying pypmc\tools\regularize_test.py -> build\lib.win-amd64-3.8\pypmc\tools
  copying pypmc\tools\_doc.py -> build\lib.win-amd64-3.8\pypmc\tools
  copying pypmc\tools\_history.py -> build\lib.win-amd64-3.8\pypmc\tools
  copying pypmc\tools\_partition.py -> build\lib.win-amd64-3.8\pypmc\tools
  copying pypmc\tools\_plot.py -> build\lib.win-amd64-3.8\pypmc\tools
  copying pypmc\tools\_probability_densities.py -> build\lib.win-amd64-3.8\pypmc\tools
  copying pypmc\tools\__init__.py -> build\lib.win-amd64-3.8\pypmc\tools
  creating build\lib.win-amd64-3.8\pypmc\tools\indicator
  copying pypmc\tools\indicator\indicator_factory_test.py -> build\lib.win-amd64-3.8\pypmc\tools\indicator
  copying pypmc\tools\indicator\_indicator_factory.py -> build\lib.win-amd64-3.8\pypmc\tools\indicator
  copying pypmc\tools\indicator\_indicator_merge.py -> build\lib.win-amd64-3.8\pypmc\tools\indicator
  copying pypmc\tools\indicator\__init__.py -> build\lib.win-amd64-3.8\pypmc\tools\indicator
  running build_ext
  creating build\temp.win-amd64-3.8
  creating build\temp.win-amd64-3.8\Release
  creating build\temp.win-amd64-3.8\Release\pypmc
  creating build\temp.win-amd64-3.8\Release\pypmc\density
  "C:\Program Files (x86)\Microsoft Visual Studio\2019\BuildTools\VC\Tools\MSVC\14.28.29910\bin\HostX86\x64\cl.exe" /c /nologo /O2 /W3 /GL /DNDEBUG /MD -IC:\Users\astiff\AppData\Local\Temp\pip-build-env-7tnf_mmt\overlay\Lib\site-packages\numpy\core\include "-Ic:\program files\python38\include" "-Ic:\program files\python38\Include" "-IC:\Program Files (x86)\Microsoft Visual Studio\2019\BuildTools\VC\Tools\MSVC\14.28.29910\include" "-IC:\Program Files (x86)\Windows Kits\10\include\10.0.19041.0\ucrt" "-IC:\Program Files (x86)\Windows Kits\10\include\10.0.19041.0\shared" "-IC:\Program Files (x86)\Windows Kits\10\include\10.0.19041.0\um" "-IC:\Program Files (x86)\Windows Kits\10\include\10.0.19041.0\winrt" "-IC:\Program Files (x86)\Windows Kits\10\include\10.0.19041.0\cppwinrt" /Tcpypmc\density\gauss.c /Fobuild\temp.win-amd64-3.8\Release\pypmc\density\gauss.obj -Wno-unused-but-set-variable -Wno-unused-function -O3
  cl : Command line error D8021 : invalid numeric argument '/Wno-unused-but-set-variable'
  error: command 'C:\\Program Files (x86)\\Microsoft Visual Studio\\2019\\BuildTools\\VC\\Tools\\MSVC\\14.28.29910\\bin\\HostX86\\x64\\cl.exe' failed with exit code 2
  ----------------------------------------
  ERROR: Failed building wheel for pypmc
Failed to build pypmc
ERROR: Could not build wheels for pypmc which use PEP 517 and cannot be installed directly

The same command gives no problems when installing on Linux (or WSL).

I believe it's just that the build is using a compiler flag that is supported by gcc, but not by Visual Studio.

see https://github.com/eos/eos/issues/411 for error messages.

After closing the issue, the same error cropped up again even though I didn't reinstall pypmc:

at 21:42:27 ❯ python3.8 -c "import pypmc"
Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "/home/fred/.local/lib/python3.8/site-packages/pypmc/__init__.py", line 9, in <module>
    from . import mix_adapt, density, sampler, tools
  File "/home/fred/.local/lib/python3.8/site-packages/pypmc/mix_adapt/__init__.py", line 7, in <module>
    from . import hierarchical, variational, pmc, r_value
  File "pypmc/mix_adapt/variational.pyx", line 1, in init pypmc.mix_adapt.variational
ValueError: numpy.ndarray size changed, may indicate binary incompatibility. Expected 88 from C header, got 80 from PyObject

Dear all

I was trying to install pypmc on my Mac (macOS 10.14.5), to use it in EOS, but the installation fails with:

pip3 install --user pypmc Collecting pypmc Using cached https://files.pythonhosted.org/packages/07/d3/86d432f20b2d82109e431c40e536ccac702a2e1cc85d6974c68b4be5d358/pypmc-1.1.2.tar.gz Requirement already satisfied: numpy in /Users/michel/Library/Python/3.7/lib/python/site-packages (from pypmc) (1.17.0) Requirement already satisfied: scipy in /Users/michel/Library/Python/3.7/lib/python/site-packages (from pypmc) (1.3.1) Building wheels for collected packages: pypmc Building wheel for pypmc (setup.py) ... error ERROR: Complete output from command /usr/local/opt/python/bin/python3.7 -u -c 'import setuptools, tokenize;__file__='"'"'/private/var/folders/4w/rfpfxc9d2ndfgz6lv9xtdrjc0000gn/T/pip-install-f1fe16hz/pypmc/setup.py'"'"';f=getattr(tokenize, '"'"'open'"'"', open)(__file__);code=f.read().replace('"'"'\r\n'"'"', '"'"'\n'"'"');f.close();exec(compile(code, __file__, '"'"'exec'"'"'))' bdist_wheel -d /private/var/folders/4w/rfpfxc9d2ndfgz6lv9xtdrjc0000gn/T/pip-wheel-esu5sr7n --python-tag cp37: ERROR: WARNING: Could not cythonize: ModuleNotFoundError("No module named 'Cython'") ... followed by a long list of errors (I can paste them here if needed). I did install cython and it is in my PATH

Any help is appreciated, thanks!