hmm-py 0.0.2

A python module to implement Hidden Markov hidden_markov for financial times series.

HMMpy

The HMMpy documentation is at https://hmmpy.readthedocs.io/en/latest/.

• Installation
• Getting started

HMMpy is a Python-embedded modeling language for hidden markov models. It currently supports training of 2-state models using either maximum-likelihood or jump estimation, and uses and API that is very similar to scikit-learn.

HMMpy began as a University project at Copenhagen Business School, where it was used for financial times series forecasting in an asset allocation project.

Installation

HMMpy is available on PyPI, and can be installed with (only for windows)

pip install hmm-py

HMMpy has the following dependencies:

• Python >= 3.8
• Cython >= 0.29
• NumPy >= 1.20.1
• Pandas >= 1.2.0
• SciPy >= 1.5.4
• tqdm

Getting started

The following code samples some data, and then trains a hidden markov model using the JumpHMM class:

from hmmpy.jump import JumpHMM
from hmmpy.sampler import SampleHMM

# Instantiate the HMM model
hmm = JumpHMM(random_state=42)

# Instantiate the sampler with user defined HMM model parameters
hmm_params = {'mu': [0.1, -0.05],
              'std': [0.1, 0.2],
              'tpm': [[1-0.0021, 0.0021],
                      [0.0120, 1-0.0120]]
             }
sampler = SampleHMM(hmm_params=hmm_params, random_state=42)

# Simulate data
observations, state_sequence = sampler.sample(n_samples=2000, n_sequences=1)  # Outputs 2000 observations and the underlying states

# Fit the model
hmm.fit(observations)

# Inspect model parameters
print(hmm.mu)
print(hmm.std)
print(hmm.tpm)