%run github_events.py


print "Some User ids from the latest events (push, star, fork etc.) on Github."
print ids[:10]
print
print "Number of unique ids found: ", ids.shape[0]
print "Largest user id: ", ids.max()

figsize(12.5, 3)
plt.hist(ids, bins=45, alpha=0.9)
plt.title("Histogram of %d Github User ids" % ids.shape[0]);
plt.xlabel("User id")
plt.ylabel("Frequency");

FACTOR = 1000000.

import pymc as pm

upper_bound = pm.Uniform("n_sign_ups", ids.max() / FACTOR, (ids.max()) / FACTOR + 1)
obs = pm.Uniform("obs", 0, upper_bound, value=ids / FACTOR, observed=True)

# code to be examplained in Chp. 3.
mcmc = pm.MCMC([upper_bound, obs])
mcmc.sample(100000, 45000)

from scipy.stats.mstats import mquantiles

samples = mcmc.trace("n_sign_ups")[:]

hist(samples, bins=100,
     label="Uniform prior",
     density=True, alpha=0.8,
     histtype="stepfilled", color="#7A68A6");

quantiles_mean = np.append(mquantiles(samples, [0.05, 0.5, 0.95]), samples.mean())
print "Quantiles: ", quantiles_mean[:3]
print "Mean: ", quantiles_mean[-1]
plt.vlines(quantiles_mean, 0, 33,
           linewidth=2, linestyles=["--", "--", "--", "-"],
           )
plt.title("Posterior distribution of total number of Github users")
plt.xlabel("number of users (in millions)")
plt.legend()
plt.xlim(ids.max() / FACTOR - 0.01, ids.max() / FACTOR + 0.12);