Source code for infpy.test.distribution_test

#
# Copyright John Reid 2006
#


import unittest, infpy, math, numpy

def checked_sample( dist ):
    """Checks the sample is in the support of the distribution"""
    x = dist.sample_from()

    # must be in support of distribution
    if not dist.supports( x ):
        raise RuntimeError( """Sampling must generate values (%d) in """
         """support of distribution (%s)""" % (x, str(type(dist))) )

    return x


class DistributionTest( unittest.TestCase ):
    """Test case for distributions"""

    def test_str( self ):
        """Test of string representation of distribution"""
        [ str( dist_gen() ) for dist_gen in infpy.Distribution.__subclasses__() ]

    def test_sample( self ):
        """Test of sample from distribution - also tests mean and variance"""
        for dist_gen in infpy.Distribution.__subclasses__():
            try:
                # create the distribution
                dist = dist_gen()

                # create some samples
                num_samples = 100000
                samples = numpy.zeros( num_samples, numpy.float64 )
                for i in xrange( num_samples ):
                    samples[i] = checked_sample( dist )

                # check statistics
                empirical_mean = numpy.mean( samples )
                theoretical_mean = dist.mean()
                theoretical_stdev = math.sqrt( dist.variance() / num_samples )
                if(
                        math.fabs( empirical_mean - theoretical_mean )
                        > 3.0 * theoretical_stdev
                ):
                    raise RuntimeError( """%s: Sample mean (%f) not within three """
                            """standard deviation (%f) of theoretical mean (%f). """
                            """# samples = %d"""
                            % (
                                    str( dist ),
                                    empirical_mean,
                                    theoretical_stdev,
                                    theoretical_mean,
                                    num_samples
                            ) )

            except Exception, detail:
                print """Problem with distribution %s: %s""" % (str(dist),detail)
                raise

    def test_pdf_derivative( self ):
        """Test of the derivative of the pdf"""
        for dist_gen in infpy.Distribution.__subclasses__():
            try:
                # create the distribution
                dist = dist_gen()

                # try a few samples
                for i in xrange( 30 ):
                    x = checked_sample( dist )

                    # check the gradients of the pdf
                    infpy.check_gradients(
                            lambda x: dist.log_pdf( x[0] ),
                            lambda x: dist.dlog_pdf_dx( x[0] ),
                            [ x ] )

            except Exception, detail:
                raise RuntimeError( """Problem with distribution %s: %s"""
                        % (str(dist),detail) )

def show_samples( dist ):
    """Test of the distribution's sampling method"""
    import pylab
    # try a few samples
    num_samples = 10000
    samples = numpy.zeros( num_samples, numpy.float64 )
    for i in xrange( num_samples ):
        samples[i] = checked_sample( dist )
    pylab.figure()
    pylab.hist( samples, bins = 100 )
    pylab.title( str( dist ) )
    pylab.show()


if __name__ == "__main__":
    # [
    #       show_samples( dist_gen() )
    #       for dist_gen
    #       in infpy.Distribution.__subclasses__()
    # ]
    # show_samples( infpy.LogNormalDistribution( 0.0, .25 ) )

    unittest.main()