Input dimension mismatch in hierarchical mixture model

[timothydmorton]

I’m fairly new to PyMC3, and I’m trying to put together a hierarchical model where I try to infer the evolution of the Dirichlet concentration parameters controlling a changing mixture of two distributions. When I try to specify the model the way that seems intuitive to me, I get a ValueError: Input dimension mis-match. I don’t understand where I need to specify shapes in order to get this to work. I share below first my data generation, and then my attempt at the model.

import numpy as np
import scipy.stats as st
import matplotlib.pyplot as plt
import pandas as pd

def linear_function(x, c0, c1):
    return c0 + c1 * (x - 0.5)


def generate_data(N=1000):
    """Generate data where mixture concentration varies
    """
    x = np.random.random(N)

    c0_0 = 0
    c1_0 = 2

    c0_1 = 0
    c1_1 = -2

    log_alpha0 = linear_function(x, c0_0, c1_0)
    log_alpha1 = linear_function(x, c0_1, c1_1)

    alpha0 = np.exp(log_alpha0)
    alpha1 = np.exp(log_alpha1)

    data = np.zeros(N)

    for i in range(N):
        weights = st.dirichlet(alpha=[alpha0[i], alpha1[i]]).rvs().squeeze()

        u = np.random.random()

        if u < weights[0]:
            data[i] = st.beta(a=3, b=1).rvs()
        else:
            data[i] = np.random.normal(3, 1)

    return pd.DataFrame({"x": x, "y": data})


data = generate_data()
data.plot.scatter("x", "y")

import pymc3 as pm
from theano import tensor as tt
import theano


with pm.Model() as model:

    x = pm.Data('x', data['x'])
    y = pm.Data('y', data['y'])

    # linear hyperparams
    c0_0 = pm.Uniform('c0_0', -10, 10)
    c1_0 = pm.Uniform('c1_0', -10, 10)

    c0_1 = pm.Uniform('c0_1', -10, 10)
    c1_1 = pm.Uniform('c1_1', -10, 10)

    
    log_alpha0 = pm.Deterministic('log_alpha0', linear_function(x, c0_0, c1_0))
    alpha0 = pm.Deterministic('alpha0', pm.math.exp(log_alpha0))                                      
    
    log_alpha1 = pm.Deterministic('log_alpha1', linear_function(x, c0_1, c1_1))
    alpha1 = pm.Deterministic('alpha1', pm.math.exp(log_alpha1))                                      
    
    alpha = pm.Deterministic('alpha', tt.stack([alpha0, alpha1]))
    
    w = pm.Dirichlet('w', a=alpha)
    
    distributions = [pm.Beta.dist(alpha=3, beta=1), pm.Normal.dist(mu=3, sigma=1)]
    
    like = pm.Mixture('like', w=w, comp_dists=distributions, observed=y)

This gives the following:

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-5-e309e2b11551> in <module>
     30     distributions = [pm.Beta.dist(alpha=3, beta=1), pm.Normal.dist(mu=3, sigma=1)]
     31 
---> 32     like = pm.Mixture('like', w=w, comp_dists=distributions, observed=y, shape=1000)
     33 
     34 #     # Trying suggestion from https://discourse.pymc.io/t/problem-with-fitting-multivariate-mixture-of-gaussians/846/2

~/miniconda3/envs/grey-matter/lib/python3.7/site-packages/pymc3/distributions/distribution.py in __new__(cls, name, *args, **kwargs)
    117         # Some distributions do not accept shape=None
    118         if has_shape or shape is not None:
--> 119             dist = cls.dist(*args, **kwargs, shape=shape)
    120         else:
    121             dist = cls.dist(*args, **kwargs)

~/miniconda3/envs/grey-matter/lib/python3.7/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 

~/miniconda3/envs/grey-matter/lib/python3.7/site-packages/pymc3/distributions/mixture.py in __init__(self, w, comp_dists, *args, **kwargs)
    155 
    156             try:
--> 157                 self.mean = (w * self._comp_means()).sum(axis=-1)
    158 
    159                 if "mean" not in defaults:

~/miniconda3/envs/grey-matter/lib/python3.7/site-packages/theano/tensor/var.py in __mul__(self, other)
    126         # and the return value in that case
    127         try:
--> 128             return theano.tensor.mul(self, other)
    129         except (NotImplementedError, TypeError):
    130             return NotImplemented

~/miniconda3/envs/grey-matter/lib/python3.7/site-packages/theano/graph/op.py in __call__(self, *inputs, **kwargs)
    251 
    252         if config.compute_test_value != "off":
--> 253             compute_test_value(node)
    254 
    255         if self.default_output is not None:

~/miniconda3/envs/grey-matter/lib/python3.7/site-packages/theano/graph/op.py in compute_test_value(node)
    128     thunk.outputs = [storage_map[v] for v in node.outputs]
    129 
--> 130     required = thunk()
    131     assert not required  # We provided all inputs
    132 

~/miniconda3/envs/grey-matter/lib/python3.7/site-packages/theano/graph/op.py in rval()
    604 
    605         def rval():
--> 606             thunk()
    607             for o in node.outputs:
    608                 compute_map[o][0] = True

~/miniconda3/envs/grey-matter/lib/python3.7/site-packages/theano/link/c/basic.py in __call__(self)
   1769                 print(self.error_storage, file=sys.stderr)
   1770                 raise
-> 1771             raise exc_value.with_traceback(exc_trace)
   1772 
   1773 

ValueError: Input dimension mis-match. (input[0].shape[1] = 1000, input[1].shape[1] = 2)

Any suggestions would be much appreciated!