ch1/1.05_decomposing_a_time_series.py at main · huncholane/time-series-cookbook

time-series-cookbook/ch1/1.05_decomposing_a_time_series.py

53 lines1.7 KBPython

1	"""
2	# Finding trend, seasonality, and remainder.
3	- Trend: Long term change.
4	- Seasonality: Repeated patterns over fixed interval such as daily.
5	- Remainder (irregular): Left from removing trend and seasonal components.
6
7	# 2 methods
8	1. Classic method: Can be multiplicative or additive where the sum/product of
9	trend, seasonality, and remainder recreate the original data
10	- Logarithmic data is the same for both
11	- Trend is estimated with a moving average
12	- Seasonality estimated by averaging values for each period
13	2. Local regression: STL/MSTL uses LOESS (locally weighted scatterplot smothing)
14	to be more flexible
15
16	Removing seasonality improves neural network forcasting performance
17	"""
18	import os
19	from statsmodels.tsa.seasonal import seasonal_decompose, STL, MSTL
20	import pandas as pd
21	import matplotlib.pyplot as plt
22
23	# Chart config
24	plt.rcParams['figure.figsize'] = (12, 6)
25
26	# Ensure assets directory
27	os.makedirs("assets", exist_ok=True)
28
29	# Load solar data
30	data = pd.read_csv(
31	"../assets/datasets/time_series_solar.csv",
32	parse_dates=["Datetime"],
33	index_col="Datetime",
34	)
35	series = data["Incoming Solar"]
36
37	# Resample to daily
38	series_daily = series.resample("D").sum()
39
40	# Perform yearly seasonal decomposition with classical method
41	result = seasonal_decompose(x=series_daily, model='additive', period=365)
42	result.plot()
43	plt.savefig("assets/classical_decomposition.png")
44
45	# Perform yearly decomposition with Seasonal Trend decomposition using LOESS (STL)
46	result = STL(endog=series_daily, period=365).fit()
47	result.plot()
48	plt.savefig("assets/stl_decomposition.png")
49
50	# Perform yearly decomposition with mutliple STL (MSTL)
51	result = MSTL(endog=series_daily, periods=(7, 365)).fit()
52	result.plot()
53	plt.savefig("assets/mstl_decomposition.png")