run_baselines_meme.py

#-*- coding: utf8

from hyperopt import fmin
from hyperopt import hp
from hyperopt import tpe

from tick.hawkes import HawkesADM4
from tick.hawkes import HawkesCumulantMatching
from tick.hawkes import HawkesSumGaussians

import numpy as np


def run_cumulants(data):
    def objective(h):
        model = HawkesCumulantMatching(h, max_iter=300)
        model.fit(data)
        return model.objective(model.adjacency)
    best = fmin(objective,
                space=hp.uniform('h', 1, 100),
                algo=tpe.suggest,
                max_evals=20)
    half_width = best['h']
    model = HawkesCumulantMatching(half_width, max_iter=300, verbose=True)
    model.fit(data)
    return model.adjacency


def run_adm4(data):
    def objective(d):
        model = HawkesADM4(d, max_iter=300, n_threads=20)
        model.fit(data)
        return -model.score()
    best = fmin(objective,
                space=hp.uniform('d', 1e-2, 1e2),
                algo=tpe.suggest,
                max_evals=20)
    decay = best['d']
    model = HawkesADM4(decay, max_iter=300, n_threads=20, verbose=True)
    model.fit(data)
    return model.adjacency


def run_sumgauss(data):
    model = HawkesSumGaussians(10, n_threads=20, max_iter=300, verbose=True)
    model.fit(data)
    return model.amplitudes.mean(axis=2)


paths = ['ticks_top100.dat', 'ticks_top500.dat']
methods = [run_adm4, run_sumgauss, run_cumulants]

for p in paths:
    realizations = []
    with open(p) as in_file:
        for line in in_file:
            spl = line.split()
            realizations.append([np.array([float(x) for x in spl[1:]])])
    for m in methods:
        name = m.__name__
        np.savez_compressed(name + '-' + p + '.npz', G=m(realizations))