Skip to content
/ MCNH-AR Public

The NHMC-AR model is a Non-Homogeneous Markov Chain AutoRegressive model. It is designed to perform context-sensitive forecasting in time series that are associated with event sequences.

License

Notifications You must be signed in to change notification settings

damaf/MCNH-AR

Repository files navigation

NHMC-AR model

Non-Homogeneous Markov Chain Auto-Regressive (NHMC-AR) model

This package contains the software programs designed for the NHMC-AR model. It includes a learning algorithm, a prediction function and a state inference algorithm.

This package has been implemented by Fatoumata Dama, PhD student (2019-2022), Nantes University, France.

Fatoumata Dama was supported by a PhD scholarship granted by the French Ministery for Higher Education, Research and Innovation. She worked under the supervision of Christine Sinoquet, Associate Professor, PhD supervisor, LS2N / UMR CNRS 6004 (Digital Science Institute of Nantes), Nantes University, France.

Requirements

  • Python 3.6
  • Numpy
  • Scipy
  • Scikit-learn
  • Pickle5
  • Futures
  • Numba 0.45
  • Tick

Anesthesia data

FC : heart frequency (HF)

PAS : systolic blood pressure (SBP)

PAM : average blood pressure (ABP)

PAD : diastolic blood pressure (DBP)

NHMC-AR model: contextual variables C1 - Application to anesthesia data

In this case, the contextual variables only take into account the latest occurrences of events, over all categories of events.

Launch model learning on anesthesia dataset

python3 -O mcnh-ar-C1_training.py train_data_dir nb_time_series model_output_dir ar_order nb_states

NHMC-AR model: contextual variables C2 - Application to anesthesia data

In this case, the contextual variables are extracted using the Hawkes point process framework. Thus, these variables take into account the whole history of past events.

Launch model learning on anesthesia dataset

python3 -O mcnh-ar-C2_training.py train_data_dir nb_time_series model_output_dir ar_order nb_states features_file

Definition of parameters

  • train_data_dir: "anesthesia_data" directory
  • nb_time_series: the number of training instances (between 1 and 500)
  • model_output_dir: the name of the directory in which model output is saved within a serialized file
  • ar_order: the order of the auto-regressive process (>= 0)
  • nb_states: the number of states to be considered (>= 2)
  • features_file: the file that contains the contextual variables C2 extracted based on the Hawkes process (to be loaded from directory "Point-process-models/tick-Hawkes-process/model_outputs/expKernel/5-event-types")

About

The NHMC-AR model is a Non-Homogeneous Markov Chain AutoRegressive model. It is designed to perform context-sensitive forecasting in time series that are associated with event sequences.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages