Following along with this example to do pretty simple bayesian linear regression using PYMC3 (learning, I hope) I get the initial example to run but then try to use my own data and get :
ValueError: Optimization error: max, logp or dlogp at max have non-finite values.
Some values may be outside of distribution support. max: {'alpha': array(50000.0),
'beta': array(50000.0), 'sigma': array(25000.0)} logp: array(nan) dlogp: array([ nan,
nan, nan])Check that 1) you don't have hierarchical parameters, these will lead to
points with infinite density. 2) your distribution logp's are properly specified.
Specific issues:
Which is suspect is due to my data range,, but it may well be that I don't understand the other parameters. Data and code is as follows: This should just run in an IPython notebook I hope. The lastqu should predict the Units, when all is said and done..
import pandas as pd
import io
content2 = '''\
Units lastqu
2000-12-31 19391 NaN
2001-12-31 35068 5925
2002-12-31 39279 8063
2003-12-31 47517 9473
2004-12-31 51439 11226
2005-12-31 59674 11667
2006-12-31 58664 14016
2007-12-31 55698 13186
2008-12-31 42235 11343
2009-12-31 40478 7867
2010-12-31 38722 8114
2011-12-31 36965 8361
2012-12-31 39132 8608
2013-12-31 43160 9016
2014-12-31 NaN 9785
'''
df2 = pd.read_table(io.BytesIO(content2))
#make sure that the columns are int, it is all a DataFrame
df2['Units']=df2['Units'][:-1].astype('int')
df2['lastqu']=df2['lastqu'][1:].astype('int')
df2
And the model code I tried is:
import pymc as pm
#import numpy as np
x=df2['lastqu'] <<<< my best guess as to how to specify my data
y=df2['Units']
trace = None
with pm.Model() as model:
alpha = pm.Normal('alpha', mu=0, sd=20)
beta = pm.Normal('beta', mu=0, sd=20)
sigma = pm.Uniform('sigma', lower=0, upper=50000)
y_est = alpha + beta * x
likelihood = pm.Normal('y', mu=y_est, sd=sigma, observed=y)
start = pm.find_MAP()
step = pm.NUTS(state=start)
trace = pm.sample(2000, step, start=start, progressbar=False)
pm.traceplot(trace);