posterior/distributions

Low-level primitives for collapsed Gibbs sampling in python and C++

distributions's Issues

Implement mixed_goodness_of_fit

see https://github.com/posterior/distributions/blob/master/distributions/util.py#L243

Implement non-normalized goodness-of-fit tests

See https://github.com/posterior/distributions/blob/master/distributions/util.py#L119

discrete_goodness_of_fit is too accepting

discrete_goodness_of_fit does not raise error when too few samples are in top 10 bins (an error that is correctly raised in density_goodness_of_fit)

Math bug in dbg.models.niw.Group.merge()

def merge(self, shared, source):
    self.count += source.count
    self.sum_x += source.sum_x
    self.sum_xxT += source.sum_xxT  # FIXME This ignores the difference in .sum_x parts.

For a possibly correct implementation in 1d, see the nich implementation.

discrete_goodness_of_fit bounds error

distributions/util.py line 113

for sample in samples:
    counts[min(int(bin_count * sample), bin_count - 1)] += 1
return multinomial_goodness_of_fit(probs, counts, len(samples), plot=plot)

protoc autogenerated files should not be in the git repository

When running make, the following error message is thrown

[  6%] Building CXX object
src/CMakeFiles/distributions_shared.dir/io/schema.pb.cc.o In file included from
/distributions/src/io/schema.pb.cc:4:0:
distributions/src/../include/distributions/io/schema.pb.h:17:2: error: #error
This file was generated by an older version of protoc which is
incompatible with your Protocol Buffer headers Please regenerate this
file with a newer version of protoc.

distributions/src/../include/distributions/io/schema.pb.h:18:2: error: #error
This file was generated by an older version of protoc which is
incompatible with your Protocol Buffer headers Please regenerate this
file with a newer version of protoc.

/distributions/src/../include/distributions/io/schema.pb.h:19:2: error: #error
This file was generated by an older version of protoc which is
incompatible with your Protocol Buffer headers Please regenerate this
file with a newer version of protoc.

make[3]: *** [src/CMakeFiles/distributions_shared.dir/io/schema.pb.cc.o] Error 1
make[3]: Leaving directory `distributions/build/python' make[2]: ***
[src/CMakeFiles/distributions_shared.dir/all] Error 2 make[2]: Leaving directory
`distributions/build/python' make[1]: *** [all] Error 2 make[1]: Leaving
directory `distributions/build/python' make: *** [python_cc] Error 2

The solution is to remove the offending file, and rebuild schema.pb.h using make protobuf:

$ rm distributions/include/distributions/io/schema.pb.h
$ make protobuf

Now running make is successful.

Sampling bug in C++ Normal Inverse Wishart

The new stronger statistical tests indicate a bug in the C++ version of Normal-Inverse-Wishart sampler. I assume the bug is in the sampling code in random.hpp (python sampler and scorer agree as per test_models.py:test_sample_value, and python and c++ scoring agree as per test_model_flavors.py:test_group, pointing to C++ sampling as the culprit.

This is now disabled in unit tests.

Failed installing 'python2 -m pip install distributions'

Cannot install distributions using command 'pip install distributions' for python2 on Ubuntu 18.04
It says /usr/bin/ld: cannot find -ldistributions_shared
error: command 'x86_64-linux-gnu-g++' failed with exit status 1

Segfault in dpd mixture

$ nosetests -v distributions/tests/test_models.py
....
distributions.tests.test_models.test_mixture_runs('lp.models.dpd',) ... Segmentation fault

$ uname -a
Linux desk 3.13.0-43-generic #72-Ubuntu SMP Mon Dec 8 19:35:06 UTC 2014 x86_64 x86_64 x86_64 GNU/Linux

Numpy deprecation test suite failure on Ubuntu 16.04

Using make all results in two failures in the distributions test suite:

======================================================================
ERROR: distributions.tests.test_util.test_scores_to_probs
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/venv/local/lib/python2.7/site-packages/nose/case.py", line 197, in runTest
    self.test(*self.arg)
  File "/posterior/distributions/distributions/tests/test_util.py", line 38, in test_scores_to_probs
    probs = scores_to_probs(scores)
  File "/posterior/distributions/distributions/util.py", line 34, in scores_to_probs
    probs = numpy.exp(scores, out=scores)
TypeError: ufunc 'exp' output (typecode 'd') could not be coerced to provided output parameter (typecode 'l') according to the casting rule ''same_kind''

======================================================================
FAIL: distributions.tests.test_models.test_add_remove('lp.models.niw',)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/venv/local/lib/python2.7/site-packages/nose/case.py", line 197, in runTest
    self.test(*self.arg)
  File "/posterior/distributions/distributions/tests/test_models.py", line 104, in test_one_model
    test_fun(module, EXAMPLE)
  File "/posterior/distributions/distributions/tests/test_models.py", line 246, in test_add_remove
    err_msg='p(x1,...,xn) != p(x1) p(x2|x1) p(xn|...)')
  File "/posterior/distributions/distributions/tests/util.py", line 120, in assert_close
    assert_less(diff, tol * norm, msg)
AssertionError: p(x1,...,xn) != p(x1) p(x2|x1) p(xn|...) off by 0.163564845877% = 0.00525343418121
-------------------- >> begin captured stdout << ---------------------
example 1/4
example 2/4
example 3/4
p(x1,...,xn) != p(x1) p(x2|x1) p(xn|...)
actual = -1.10329115391
expected = -1.10854458809

--------------------- >> end captured stdout << ----------------------

----------------------------------------------------------------------
Ran 292 tests in 55.957s

FAILED (SKIP=15, errors=1, failures=1)
make: *** [test_cy] Error 1

The numpy error in distributions.tests.test_util.test_scores_to_probs is due to deprecation: numpy/numpy#6464

Not sure why the probabilities are failing.