 |
 |
|
|
Lab 9: Time Series - Motif Discovery
In [1]:
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import ts_functions as ts
data = pd.read_csv('data/ashrae_single.csv')
index_var = 'timestamp'
variable = 'meter_reading'
data[index_var] = pd.to_datetime(data[index_var])
data = data.set_index(index_var).sort_index()
FIG_WIDTH, FIG_HEIGHT = 3*ts.HEIGHT, ts.HEIGHT/2
plt.figure(figsize=(FIG_WIDTH, FIG_HEIGHT))
ts.plot_series(data, x_label='timestamp', y_label='consumption', title='ASHRAE original')
plt.xticks(rotation = 45)
plt.show()
Matrix Profile
Compute Matrix Profiles
Here we compute Matrix Profiles over varying window sizes:8 hours Half of a day (12 hours) A day (24 hours) A week (7 days) A month (30 days) A quarter (90 days)
In [2]:
import matrixprofile as mp
all_windows = [
('8 Hours', 8),
('12 Hours', 12),
('Day', 24),
('Week', 7 * 24),
('Month', 30 * 24),
('Quarter', 3 * 30 * 24),
]
def compute_matrix_profiles(df: pd.DataFrame, windows: list) :
profiles = {}
for label, size in windows:
key = '{} Profile'.format(label)
profiles[key] = mp.compute(df[variable].values, size)
return profiles
def plot_signal_data(profiles: dict, windows: list):
_, axes = plt.subplots(len(windows), 1, sharex=True, figsize=(FIG_WIDTH, len(windows)*FIG_HEIGHT))
for ax_idx, window in enumerate(windows):
key = '{} Profile'.format(window[0])
axes[ax_idx].plot(profiles[key]['mp'])
axes[ax_idx].set_title(key)
plt.xlabel(index_var)
plt.tight_layout()
plt.show()
all_profiles = compute_matrix_profiles(data, all_windows)
plot_signal_data(all_profiles, all_windows)
Motifs
In [3]:
def compute_all_profiles(profiles: dict, windows: list, k: int, type: str='motifs'):
discover_function = mp.discover.motifs
if type == 'discords':
discover_function = mp.discover.discords
for label, size in windows:
key = '{} Profile'.format(label)
profiles[key] = discover_function(profiles[key], k=k, exclusion_zone=2*size//3)
compute_all_profiles(all_profiles, all_windows, k=5, type='motifs')
In [4]:
def show_profile(profile, title, type):
lst_figs = mp.visualize(profile)
for i in range(len(lst_figs)-1):
plt.close(lst_figs[i])
plt.suptitle(type + ' - ' + title)
plt.show()
title = all_windows[0][0]+' Profile'
show_profile(all_profiles[title], title, 'Motifs')
In [5]:
title = all_windows[2][0]+' Profile'
show_profile(all_profiles[title], title, 'Motifs')
In [6]:
title = all_windows[3][0]+' Profile'
show_profile(all_profiles[title], title, 'Motifs')
In [7]:
title = all_windows[4][0]+' Profile'
show_profile(all_profiles[title], title, 'Motifs')
In [8]:
title = all_windows[5][0]+' Profile'
show_profile(all_profiles[title], title, 'Motifs')
Discords / Anomalies
In [9]:
compute_all_profiles(all_profiles, all_windows, k=5, type='discords')
In [10]:
title = all_windows[0][0]+' Profile'
show_profile(all_profiles[title], title, 'Discords')
In [11]:
title = all_windows[3][0]+' Profile'
show_profile(all_profiles[title], title, 'Discords')
 |
 |
 |