PyMC MCMC analysis - .summary() "

Question

I'm new to PyMC and am having a little trouble getting out parameters related to my prior, for example, the mean and standard deviation.

I describe my model in a file called 'model.py' like so:

    import pymc
    import numpy

    #constants
    r_div=numpy.loadtxt("r_div", comments="#", unpack=False)
    map=numpy.loadtxt("map", comments="#", delimiter=",", unpack=False)
    M_star=3*10^6;

    #data
    n=numpy.loadtxt("n")

    #priors
    alpha_0=pymc.Uniform('alpha_0end',-10,10, value=0)
    logA_0=pymc.Uniform('logA_0end',-10,10,value=-6.1246)

    #model

    @pymc.deterministic(plot=False)

    def r(logA_0=logA_0,alpha_0=alpha_0,M_star=M_star,r_div=r_div):

        r=r_div*numpy.exp(logA_0)*((numpy.exp(map[:,1])/M_star)**(alpha_0))

        return r

            #likelihood
            Distribution=pymc.Poisson('Distribution',mu=r,value=n,observed=True)

And then I use the following script in ipython to run the MCMC chain:

            import pymc
            import model
            M=pymc.MCMC(model)
            M.sample(100000, burn=10000)
            M.summary()

Everything seems to work until the final command M.summary() where I get the error:

AttributeError Traceback (most recent call last) in () ----> 1 M.summary()

AttributeError: 'MCMC' object has no attribute 'summary'

I am confident the chain has run successfully as the command M.trace('alpha_0end')[:] shows there are chain elements there but I can't get out any information out about the prior, such as a mean or standard deviation. I've tried different permutations of the summary command. For example: M.alpha_0end.summary()

It would be helpful to know if there is an easy way to get out the standard deviation and means of the priors.

score 0 · Accepted Answer · answered Jun 17 '14 at 19:32

I can't run your code, but M.summary() works for me in this minimal example:

In [1]: import pymc as pm

In [2]: pm.__version__
Out[2]: '2.3.2'

In [3]: X = pm.Normal('X', 0, 1)

In [4]: M = pm.MCMC(dict(X=X))

In [5]: M.sample(100000, burn=10000)
 [-----------------100%-----------------] 100000 of 100000 complete in 5.0 sec
In [6]: M.summary()

    X:

        Mean             SD               MC Error        95% HPD interval
        ------------------------------------------------------------------
        -0.0             1.003            0.003            [-1.897  2.026]


        Posterior quantiles:

        2.5             25              50              75             97.5
         |---------------|===============|===============|---------------|
        -1.961           -0.673          0.0            0.675         1.964

score 0 · Answer 2 · answered Jul 10 '14 at 22:00

The summary subroutine was added in February 2012 (a04551094a70451a4afee4e11be8b6872e74fe46). At that time, the version number was 2.2-grad.

If your pm.__version__ is 2.2-grad or lower, you have a version of pymc that is too old to use this function. Version 2.2-beta or later should work.

PyMC MCMC analysis - .summary() "

2 Answers2