scikit-learn GaussianHMM ValueError: input must be a square array

Question

I am working with scikit-learn's GaussianHMM and am getting the following ValueError when I try to fit it to some observations. here is code that demonstrates the error:

>>> from sklearn.hmm import GaussianHMM
>>> arr = np.matrix([[1, 2, 3], [4, 5, 6], [7, 8, 9]])
>>> arr
matrix([[1, 2, 3],
        [4, 5, 6],
        [7, 8, 9]])
>>> gmm = GaussianHMM ()
>>> gmm.fit (arr)
/System/Library/Frameworks/Python.framework/Versions/2.7/Extras/lib/python/numpy/lib/function_base.py:2005: RuntimeWarning: invalid value encountered in divide
  return (dot(X, X.T.conj()) / fact).squeeze()
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/Library/Python/2.7/site-packages/sklearn/hmm.py", line 427, in fit
    framelogprob = self._compute_log_likelihood(seq)
  File "/Library/Python/2.7/site-packages/sklearn/hmm.py", line 737, in _compute_log_likelihood
    obs, self._means_, self._covars_, self._covariance_type)
  File "/Library/Python/2.7/site-packages/sklearn/mixture/gmm.py", line 58, in log_multivariate_normal_density
    X, means, covars)
  File "/Library/Python/2.7/site-packages/sklearn/mixture/gmm.py", line 564, in _log_multivariate_normal_density_diag
    + np.dot(X ** 2, (1.0 / covars).T))
  File "/System/Library/Frameworks/Python.framework/Versions/2.7/Extras/lib/python/numpy/matrixlib/defmatrix.py", line 343, in __pow__
    return matrix_power(self, other)
  File "/System/Library/Frameworks/Python.framework/Versions/2.7/Extras/lib/python/numpy/matrixlib/defmatrix.py", line 160, in matrix_power
    raise ValueError("input must be a square array")
ValueError: input must be a square array
>>>

How might I remedy this? It seems that I am giving it valid inputs. Thanks!

If i make arr such that it is not a square matrix, however, I still get the same error even if i encapsulate it in brackets... i.e. if arr is matrix([[ 1, 2, 3], [ 4, 5, 6], [ 7, 8, 9], [10, 11, 12]]). Any ideas? Thanks! — Jay Hack, Dec 17 '13 at 03:57

score 3 · Answer 1 · answered Dec 16 '13 at 20:03

You have to fit with a list, see official examples:

>>> gmm.fit([arr])
GaussianHMM(algorithm='viterbi', covariance_type='diag', covars_prior=0.01,
      covars_weight=1,
      init_params='abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ',
      means_prior=None, means_weight=0, n_components=1, n_iter=10,
      params='abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ',
      random_state=None, startprob=None, startprob_prior=1.0, thresh=0.01,
      transmat=None, transmat_prior=1.0)
>>> gmm.n_features
3
>>> gmm.n_components
1

The link is dead. See update here. http://scikit-learn.org/0.14/auto_examples/applications/plot_hmm_stock_analysis.html — Fanglin, Aug 15 '14 at 23:13

unutbu · Answer 2 · 2016-06-24T21:55:32.420

3

According to the docs, gmm.fit(obs) expects obs to be a list of array-like objects:

obs : list
    List of array-like observation sequences (shape (n_i, n_features)).

Therefore, try:

import numpy as np
from sklearn.hmm import GaussianHMM
arr = np.matrix([[1, 2, 3], [4, 5, 6], [7, 8, 9]])
gmm = GaussianHMM()
print(gmm.fit([arr]))

Hidden markov models (HMMs) are no longer supported by sklearn.

edited Jun 24 '16 at 21:55

answered Dec 16 '13 at 20:04

unutbu

842,883
184
1,785
1,677

so what I'm unclear on is what is represented in each column if we have to have as many columns as emission possibilities? Why can't we just pass a 1-dimensional sequence of emissions? – Brooks Jun 11 '15 at 20:21

scikit-learn GaussianHMM ValueError: input must be a square array

2 Answers2

Linked