ruizhang2016 / efficient-nonparametric-bayesian-hawkes-processes

Efficient-Nonparametric-Bayesian-Hawkes-Processes

Implementation of Efficient Non-parametric Bayesian Hawkes Processes in Python3.5. A tutorial is included. There is no implementation of Halpin's trick which will be uploaded.

Required Packages

• numpy
• scipy==1.2.0
• matplotlib
• autograd==1.1.13
• tick==0.5.0.0
• numpydoc==0.7.0

They can be installed through pip:

   $ pip3 install numpy scipy==1.2.0 matplotlib autograd==1.1.13 tick==0.5.0.0 numpydoc==0.7.0

Citation

If you find Efficient Nonparametric Bayesian Hawkes Processes useful in your research, please consider citing:

@article{zhang2019efficient,
	title={Efficient Non-parametric Bayesian Hawkes Processes},
	author={Zhang, Rui and Walder, Christian and Rizoiu, Marian Andrei and Xie, Lexing},
	journal={the 28th International Joint Conference on Artificial Intelligence},
	year={2019}
}

Tutorial

See Gibbs_Hawkes.ipynb.

Online Demo

An online demo is on GoogleColab.

License

MIT License

efficient-nonparametric-bayesian-hawkes-processes's Issues

Just a small typo in your bibTeX

@Article{zhang2019efficient,
title={Efficient Non-parametric Bayesian Hawkes Processes},
author={Zhang, Rui and Walder, Christian and Rizoiu, Marian Andrei and Xie, Lexing},
journal={the 28th International Joint Conference on Artificial Intelligence},
year={2019}
}

I think this should be zhang2018efficient in your README page. Please double-check with Google Scholar.
I just came across this repo when checking on zhang2019efficient. Have a good day.

Error while running on different data

Hello,

I tried to use your method on simulated events from a univariate and linear Hawkes process with different kernels that the cosine function present in the example from Gibbs_Hawkes.ipynb.

No matter the kernel I choose to simulate, I get the same mistake when inferring saying that Qinv is not definite positive while running. Here is an example:

end_time = 10
baseline = np.array([.2])
alpha = np.array([[0.8]])
kernel = 'expon'
random_state = 27
events = simu_hawkes_cluster(end_time, baseline, alpha, 
                             kernel, random_state=random_state)
hp_data_list = [events[0].reshape(1, -1)]
xplot = np.linspace(0., 1, 100).reshape((1, -1))
rm_f, rstd_f, rm_mu, rstd_mu = gh.infer(hp_data_list, xplot, num_ite=800, burn_in=500)

Note that the function simu_hawkes_cluster is a custom function to simulate Hawkes process. The error returned is the following:

Thanks in advance!

Halpin's trick

Has this been implemented?

If not, then can I assume the infer() function is quadratic in complexity instead of linear?

Thank you,

-Richard