When I run the example code in Section 10.2.5 of the documentation at http://alkaline-ml.com/pmdarima/usecases/sun-spots.html the best fitting model I obtain is ARIMA(0,1,2)(0,0,0)[12] but this is not the one that should be obtained. Further inconsistencies arise with the results in the documentation e.g. the forecast values are constant.
To Reproduce
Steps to reproduce the behavior:
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import pmdarima as pm
from pmdarima.datasets import load_sunspots
from pmdarima.model_selection import train_test_split
from pmdarima.preprocessing import BoxCoxEndogTransformer
y = load_sunspots(True)
train_len = 2750
y_train, y_test = train_test_split(y, train_size=train_len)
y_train.head()
from pmdarima.pipeline import Pipeline
fit2 = Pipeline([
('boxcox', BoxCoxEndogTransformer(lmbda2=1e-6)),
('arima', pm.AutoARIMA(trace=True,
suppress_warnings=True,
m=12))
])
# Fit the model
fit2.fit(y_train)
# Predict next 70 values
fit2.predict(70)
Expected behavior
I'd expect the model identified to be consistent with that in the documentation, namely SARIMAX(3, 1, 2), and for the predicted values to be non-constant.
Describe the bug
When I run the example code in Section 10.2.5 of the documentation at http://alkaline-ml.com/pmdarima/usecases/sun-spots.html the best fitting model I obtain is ARIMA(0,1,2)(0,0,0)[12] but this is not the one that should be obtained. Further inconsistencies arise with the results in the documentation e.g. the forecast values are constant.
To Reproduce
Steps to reproduce the behavior:
Versions
Expected behavior I'd expect the model identified to be consistent with that in the documentation, namely SARIMAX(3, 1, 2), and for the predicted values to be non-constant.
Actual behavior
Here is the output showing the search path:
And here are the predicted values:
Additional context