Hi,

while reading your code, I wonder is there a requirement of the emission distribution in the model fitting? If I know the emission distribution is gaussian, how the model fitting will be? Does it will fit to such specific distribution or not? It seems to me that there's no such requirement.

Just want to make sure.

Hi,

after installing hsmmlearn, I ran "python3 -m unittest discover -v ."

I got: ImportError: No module named 'hsmmlearn.base".

Any idea why? Thanks.

I installed hsmmlearn as described and the installation went through. However, when I run the tests, I get

======================================================================
ERROR: hsmmlearn.tests.test_base (unittest.loader._FailedTest)
----------------------------------------------------------------------
ImportError: Failed to import test module: hsmmlearn.tests.test_base
Traceback (most recent call last):
  File "/home/ano/anaconda3/lib/python3.6/unittest/loader.py", line 428, in _find_test_path
    module = self._get_module_from_name(name)
  File "/home/ano/anaconda3/lib/python3.6/unittest/loader.py", line 369, in _get_module_from_name
    __import__(name)
  File "/home/ano/hsmmlearn/hsmmlearn/tests/test_base.py", line 7, in <module>
    from hsmmlearn.base import _viterbi_impl
ImportError: /home/ano/hsmmlearn/hsmmlearn/base.cpython-36m-x86_64-linux-gnu.so: undefined symbol: __cxa_throw_bad_array_new_length

The same ImportError occurs when I try to import classes such as e.g. GaussianHSMM:

---------------------------------------------------------------------------
ImportError                               Traceback (most recent call last)
<ipython-input-9-afdd3f146f00> in <module>()
----> 1 from hsmmlearn.hsmm import MultinomialHSMM

~/anaconda3/lib/python3.6/site-packages/hsmmlearn-0.1.0-py3.6-linux-x86_64.egg/hsmmlearn/hsmm.py in <module>()
      1 import numpy as np
      2 
----> 3 from .base import _viterbi_impl, _fb_impl
      4 from .emissions import GaussianEmissions, MultinomialEmissions
      5 from .properties import Durations, Emissions, TransitionMatrix

ImportError: /home/ano/anaconda3/lib/python3.6/site-packages/hsmmlearn-0.1.0-py3.6-linux-x86_64.egg/hsmmlearn/base.cpython-36m-x86_64-linux-gnu.so: undefined symbol: __cxa_throw_bad_array_new_length

I am running Python 3.6 and GCC 7.3.1-2 on Fedora 27. Any ideas what I have to change about my build configuration to make this work?

Apart from that: Thanks for the great work!

Hello, Could you tell me which paper you referred to to write this algorithm? I have some questions about the EM algorithm.

Given the sequence of observations, we use the EM algorithm to estimate the parameters of the HSMM model. In the EM algorithm, we first need to give the initial value, and then iterate through the EM algorithm to get the optimal value.

The first step: we use the following code, this code gives the real value of tmat, means, scales, durations and other HSMM parameters to get the observation sequence and hidden sequence.

The second step: use the EM algorithm to solve the parameters of the HSMM model. Firstly, give the hypothetical initial values of the new_tmat, new_means, new_scales, and new_durations parameters, and then use the hsmm.fit function to train our observation sequence to obtain the parameters of the HSMM. Are the parameters obtained in this step approximate to the real values?

I used the following initial dummy parameters and handcrafted a super toy dataset, then use the model to fit. However, the posterior has negative probability values.

#assume three hidden states
prior = np.array([[0.1,0.1,0.1,0.7],
                  [0.1,0.1,0.7,0.1],
                  [0.1,0.7,0.1,0.1]])
durations = np.array([[0.1, 0.1, 0.8, 0.0, 0.0],
                      [0.1, 0.7, 0.2, 0.0, 0.0],
                      [0.2, 0.2, 0.2, 0.2, 0.2]])
tmat = np.array([[0.0, 0.5, 0.5],
                 [0.3, 0.0, 0.7],
                 [0.6, 0.4, 0.0]])

Toy data:

D = 5
samples = list()
for i in range(3): # simply repeat the pattern 3 times.
    for s in range(4): # 4 categories
        samples.append([s for _ in range(D)]) # duration of each categories are always D
samples = np.array(tmp).reshape(-1)

Fit:

new_hsmm = MultinomialHSMM(
    prior, durations, tmat,
)
new_hsmm.fit(samples)

Result:

print new_hsmm.emissions._probabilities
[[  5.00000000e-01  -3.08395409e-16  -3.08395358e-16   5.00000000e-01]
 [  2.50000000e-01   6.94826100e-17   5.00000000e-01   2.50000000e-01]
 [  7.22089011e-16   8.33333333e-01   1.66666667e-01   3.03954851e-14]]

Hi, I am trying to find the API documentation reference. I google it and found the following link

http://hsmmlearn.readthedocs.io/en/latest/source/reference.html

But the API reference is empty. Can you point to me the right link?

Thanks,

Ben

I write to switch system to poetry.

Because, poetry add "git+https://github.com/jvkersch/hsmmlearn" is failed. poetry don't load system Cython.

Poetry doesn`t use provided private repo when compiling dependencies from source · Issue #3744 · python-poetry/poetry

2021-06-02T18:46:48 ✖ poetry add "git+https://github.com/jvkersch/hsmmlearn"

  PackageInfoError

  Unable to determine package info for path: /tmp/pypoetry-git-hsmmlearn0klkxnrl

  Fallback egg_info generation failed.

  Command ['/tmp/tmpqwo321rq/.venv/bin/python', 'setup.py', 'egg_info'] errored with the following return code 1, and output:
  Traceback (most recent call last):
    File "/tmp/pypoetry-git-hsmmlearn0klkxnrl/setup.py", line 54, in <module>
      ext_modules=get_extension_modules(),
    File "/tmp/pypoetry-git-hsmmlearn0klkxnrl/setup.py", line 23, in get_extension_modules
      from Cython.Build import cythonize
  ModuleNotFoundError: No module named 'Cython'

  at ~/.local/share/pypoetry/venv/lib/python3.9/site-packages/poetry/inspection/info.py:502 in _pep517_metadata
      498│                 try:
      499│                     venv.run("python", "setup.py", "egg_info")
      500│                     return cls.from_metadata(path)
      501│                 except EnvCommandError as fbe:
    → 502│                     raise PackageInfoError(
      503│                         path, "Fallback egg_info generation failed.", fbe
      504│                     )
      505│                 finally:
      506│                     os.chdir(cwd.as_posix())

Even if you were to upload it to PyPi. Basically, it is better to keep the information needed for the build closed locally. It is easier to keep the information needed to build locally, even if it is uploaded to PyPi, so I let poetry do the work and made the global library as unnecessary as possible.

This is a near draft, it still has data from the previous build system, and it doesn't properly accept information from the previous build system. If this project is still alive and accepting PRs, I will work on it more carefully.

I have cloned the git repository and I think I have all the required files. But I can't import anything inside hsmmlearn. Also. print(hsmmlearn.file) gives 'none'. What should I do?

Could you please give a hint, how I can fit model parameters using high dimensional data. For example if I have Data(n*p) n-> number of observation and p -> dimension of observation.

We are working on device failure detection using syslogs and trying to use HSMM Model for the same .We have built training data for Error sequences and another with Errorfree sequences. We want to build a HSMM model for both Error and Error free data using the hsmmlearn module.In the example model demonstrated in the git, model is trained on only the sample data based out of Gaussian distribution, but we require the HSMM model to be trained on our Error and Error free data.Kindly help in understanding of how to fit hsmmlearn model for our training data(Error and Error Free data) and do predictions.

I want to get probability for a given observation sequence. I could see liklihood function is there in emission package, but that is not giving one value.

new_hsmm.emissions.likelihood(np.array([1,2,3])) Ouput array([[2.26e-01, 4.26e-02, 2.76e-03], [4.25e-04, 8.43e-03, 7.06e-02], [3.41e-32, 6.83e-26, 2.67e-20]])

I want one single value. Is it possible ?

Thanks