I want to make a mixture of two TruncatedNormal distributions in pymc3.
I am trying to modify this piece of documentation. See example #2 for Poisson.
import pymc3 as pm
import numpy as np
x_obs = [x for x in np.random.randn(250) if 1>x>0] + [x for x in np.random.randn(250) if 1.2>x>0.2]
x = np.array(x_obs)
with pm.Model() as model:
w = pm.Dirichlet("w", [1,1])
mu1 = pm.TruncatedNormal("mu1", mu=x.mean(), sigma=1, lower=0, upper=1)
tau1 = pm.Gamma("tau1", mu=1/x.std()**2, sigma=1.0)
mu2 = pm.TruncatedNormal("mu2", mu=x.mean(), sigma=1, lower=0, upper=1)
tau2 = pm.Gamma("tau2", mu=1/x.std()**2, sigma=1.0)
tn1 = pm.TruncatedNormal.dist("tn1",
mu=mu1,
tau=tau1,
lower=0,
upper=1)
tn2 = pm.TruncatedNormal.dist("tn2",
mu=mu2,
tau=tau2,
lower=0,
upper=1)
like = pm.Mixture(name="like",
w=w,
comp_dists=[tn1, tn2],
observed = x)
The error I am getting is:
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
<ipython-input-7-0287db35d04c> in <module>
5 # mu2 = pm.TruncatedNormal("mu2", mu=x.mean(), sigma=1, lower=0, upper=1)
6 # tau2 = pm.Gamma("tau2", mu=1/x.std()**2, sigma=1.0)
----> 7 tn1 = pm.TruncatedNormal.dist("tn1",
8 mu=mu1,
9 tau=tau1,
~/.conda/envs/user_env/lib/python3.9/site-packages/pymc3/distributions/distribution.py in dist(cls, *args, **kwargs)
128 def dist(cls, *args, **kwargs):
129 dist = object.__new__(cls)
--> 130 dist.__init__(*args, **kwargs)
131 return dist
132
TypeError: __init__() got multiple values for argument 'mu'
What is the correct way to apply mixture models for truncated normals?
You don't need to give the distribution a name. The string tn1 passed as first the argument to pm.TruncatedNormal.dist is interpreted as mu (as you pass a named mu also, you get the exception). Try
tn1 = pm.TruncatedNormal.dist(mu=mu1, tau=tau1, lower=0, upper=1)
(and tn2 accordingly).