add a flexible distribution set of classes

[segasai]

Here is the infrastructure to create any arbitrary distribution from segments of normal distributions.

Here is how you define a Chabrier2005-like IMF dd=distributions.CompositeDistribution([distributions.TruncatedLogNormal(0.3,0.5,0.08,1),distributions.PowerLaw(-2.55,1,np.inf)])

The distributions support .pdf() .cdf() and .rvs() methods to generate samples. All the distributions -- are proper probability distributions (not densities) so they are all by construction integrate to 1.

My suggestion, is to replace piece by piece the imf.py code by this.

[keflavich]

:+1: thanks! I don't suppose you have any tests already built for this? I'll go ahead and merge, but it would be great if we had some correctness / consistency tests.

[segasai]

No I didn't. How do yo run your tests ? I wasn't able to run them due to 'from .. import imf' lines

[keflavich]

Either cd to the source directory and run pytest (this is the automated framework) or, with imf installed, do something like:

python -c "import imf.tests.test_imf; imf.tests.test_imf.test_kroupa_integral()"

[segasai]

That's exactly what I tried before and it didn't work for me I'm using python 3.5

__________________ ERROR collecting imf/tests/test_imf.py ____________________
imf/tests/test_imf.py:4: in <module>
    from .. import imf
E   SystemError: Parent module '' not loaded, cannot perform relative import

I think something changed in relative imports in python, that I don't fully understand

[keflavich]

That's my mistake. In order for tests/ to be a module, a (blank) __init__.py file is required. I had created one locally but forgot to push it to github. Grab the latest master and try again?

[segasai]

thanks, it works now

[segasai]

I've put a new pull request with rudimentary tests