The hiccups¶
As a researcher I am interested in detecting hiccup, but need more date so I want to generate some time series. The scenario I want to model is
- Hiccups should be observed for a week, except during sleeping (23:00 - 8:00).
- A hiccup happens once an hour at a random time
- Time series should have a time point every minute
Other scenarios that would nice
- Multiple hiccups occur in a bout (see https://en.wikipedia.org/wiki/Hiccup for histogram)
- Person has infrequent multiple hiccups once a day
- Person has higher chance of hiccup after eating (from 8:00 - 9:00, 12:00 - 13:00 and 18:00 - 19:00)
In [1]:
from math import floor
import matplotlib.pyplot as plt
import numpy as np
from sequgen.deterministic.triangular_peak import triangular_peak
from sequgen.dimension import Dimension
from sequgen.parameter_space import ParameterSpace
In [2]:
timepoints = (np.concatenate((
np.arange('2020-03-01T08:00', '2020-03-01T23:00', dtype='datetime64[m]'),
np.arange('2020-03-02T08:00', '2020-03-02T23:00', dtype='datetime64[m]'),
np.arange('2020-03-03T08:00', '2020-03-03T23:00', dtype='datetime64[m]'),
np.arange('2020-03-04T08:00', '2020-03-04T23:00', dtype='datetime64[m]'),
np.arange('2020-03-05T08:00', '2020-03-05T23:00', dtype='datetime64[m]'),
)) - np.datetime64('2020-03-01T08:00')).astype(int) # Convert datatimes into minutes since 2020-03-01T08:00
In [3]:
space = ParameterSpace((
Dimension('placement', 0, 59), # minute in an hour
Dimension('height', 5, 10), # severity
))
hiccups = np.zeros(len(timepoints), dtype=float)
for day in range(5):
daily_offset = day * (24 * 60)
for hour in range(15):
parameters = space.sample()
# add offset, watchout placement should pick a value from timepoints, not an index
parameters['placement'] = floor(parameters['placement']) + daily_offset + hour * 60
hiccup = triangular_peak(timepoints, width_base_left=0, width_base_right=2, **parameters) # TODO unable to make peak of 1 time point
hiccups += hiccup
In [6]:
def plot(title, x, y):
plt.figure(figsize=(20, 4))
plt.plot(x, y, ".b-")
plt.title(title)
plt.grid(True)
plt.show()
plot('Hiccups', timepoints.astype('timedelta64[m]') + np.datetime64('2020-03-01T08:00'), hiccups)
In [ ]: