In [ ]:

    

import astropy.units as u
import astropy.coordinates as coord
import numpy as np
from numpy.polynomial.polynomial import polyval
import matplotlib.pyplot as pl
pl.style.use('apw-notebook')
%matplotlib inline
import h5py
from scipy.ndimage import gaussian_filter
from scipy.stats import scoreatpercentile

from astroML.density_estimation import XDGMM
from astroML.utils import log_multivariate_gaussian
from astroML.plotting.tools import draw_ellipse
from scipy.misc import logsumexp


    

In [ ]:

    

XCov_filename = "/Users/adrian/projects/globber/data/ngc5897/XCov_med.h5"


    

In [ ]:

    

with h5py.File(XCov_filename, "r") as f:
    cluster_X = f['cluster']['X'][:]
    cluster_Cov = f['cluster']['Cov'][:]


    

In [ ]:

    

cluster_X.shape


    

In [ ]:

    

clf = XDGMM(n_components=16, verbose=True, n_iter=128)


    

In [ ]:

    

clf.fit(cluster_X, cluster_Cov)


    

In [ ]:

    

sample_X = clf.sample(size=4096)


    

In [ ]:

    

fig,axes = pl.subplots(1,3,figsize=(12,6),sharex=True,sharey=True)

axes[0].plot(cluster_X[:,1], cluster_X[:,0], ls='none', alpha=0.5)
axes[1].plot(sample_X[:,1], sample_X[:,0], ls='none', alpha=0.5)

for i in range(clf.n_components):
    draw_ellipse(clf.mu[i,:2][::-1], clf.V[i,:2,:2][::-1,::-1], scales=[2],
                 ec='k', fc='gray', alpha=0.2, ax=axes[2])

pl.xlim(0,0.7)
pl.ylim(22,14)


    

In [ ]:

    

import pickle


    

In [ ]:

    

with open("/Users/adrian/projects/globber/data/ngc5897/xd_trained.pickle", 'wb') as f:
    pickle.dump(clf, f)


    

In [ ]:

    

logprob = clf.logprob_a(cluster_X, cluster_Cov)


    

In [ ]:

    

logprob.shape


    

In [ ]:

    

logsumexp(self.logprob_a(X, Xerr), -1)