Neural network-based point process models
Synthetic point processes are simulated using tick
and point-process-rust
.
Neural networks are written using PyTorch.
Objective
Use a modified neural network-based Hawkes process model for next event prediction in a time series.
Overview
Directories:
data
contains the databiblio
contains the bibliographyrapport
contains the project reportnotebooks
contains the Jupyter notebooks
Installation
Loading a model
From Saving & Loading models.
If the model state dict was saved with torch.save(model.state_dict(), PATH)
import torch
from models import ModelClass
model = ModelClass(*args, **kwargs)
model.load_state_dict(torch.load(PATH))
model.eval() # Evaluation mode
References
- Second order statistics characterization of Hawkes processes and non-parametric estimation (arXiv :1401.0903) E. Bacry, J.F. Muzy. Trans. in Inf. Theory, 62, Iss.4 (2016) https://arxiv.org/abs/1401.0903
- Estimation of slowly decreasing Hawkes kernels : Application to high frequency order book modelling (arXiv :1412.7096) E.Bacry, T.Jaisson, J.-F.Muzy Quantitative Finance Vol.16 Iss. 8 (2016) https://arxiv.org/abs/1412.7096
- Hawkes processes in finance. (arXiv :1502.04592) E.Bacry, I.Mastromatteo, J.-F.Muzy Market Microstructure and Liquidity Vol. 01, No. 01, 1550005 (2015). https://arxiv.org/abs/1502.04592
- The Neural Hawkes Process: A Neurally Self-Modulating Multivariate Point Process, H. Mei, J. Eisner https://arxiv.org/pdf/1612.09328.pdf
- Modeling The Intensity Function Of Point Process Via Recurrent Neural Networks https://arxiv.org/pdf/1705.08982.pdf
- On a Bayesian RNN for learning the decrease speed parameter in a process: Neural Hawkes Process Memory (Mike Mozer) http://www.cs.colorado.edu/~mozer/Research/Selected%20Publications/talks/Mozer_NeuralHawkesProcessMemory_NIPS2016.pdf
- Recurrent Marked Temporal Point Processes: https://www.kdd.org/kdd2016/papers/files/rpp1081-duA.pdf
- Numerical Recipes The Art of scientific Computing (An amazing book for not analytic integration) https://e-maxx.ru/bookz/files/numerical_recipes.pdf