pmdarima_auto_arima_parameter_discovery_and_forecast.py

1import pmdarima as pm
2from pmdarima.model_selection import train_test_split
3import numpy as np
4import matplotlib.pyplot as plt
5
6# Load the data and split it into training and test sets
7data = pm.datasets.load_wineind()
8train, test = train_test_split(data, train_size=150)
9
10# Fit a simple auto_arima model
11model = pm.auto_arima(train, start_p=1, start_q=1,
12                      test='adf',       # use adftest to find optimal 'd'
13                      max_p=3, max_q=3, # maximum p and q
14                      m=1,              # frequency of series
15                      d=None,           # let model determine 'd'
16                      seasonal=False,   # No Seasonality
17                      start_P=0, 
18                      D=0, 
19                      trace=True,
20                      error_action='ignore',  
21                      suppress_warnings=True, 
22                      stepwise=True)
23
24print(model.summary())
25
26# Make your forecasts
27forecasts = model.predict(n_periods=test.shape[0])  # predict N steps into the future
28
29# Visualize the forecasts (optional)
30x = np.arange(data.shape[0])
31plt.plot(x[:150], train, c='blue')
32plt.plot(x[150:], forecasts, c='green')
33plt.show()