baselines.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


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']
    print(decay)
    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.baselines.mean(axis=2)