<aside> 💡 Auto select ARIMA p, d, q

</aside>

Auto-ARIMA - Seasonal(12)

Raw Data

model = pm.auto_arima(
  train['Passengers'],
  trace=True,
  suppress_warnings=True,
  seasonal=True,
  m=12
)
model.summary()

Log Data

logmodel = pm.auto_arima(
  train['LogPassengers'],
  trace=True,
  suppress_warnings=True,
  seasonal=True,
  m=12
)
logmodel.summary()

Auto-ARIMA - no seasonal

Stepwise

### non-seasonal, stepwise=True
model = pm.auto_arima(
  train['LogPassengers'],
  trace=True,
  suppress_warnings=True,
  max_p=12, max_q=2, max_order=14,
  stepwise=True,
  seasonal=False
)

Grid

### non-seasonal
model = pm.auto_arima(
  train['LogPassengers'],
  trace=True,
  suppress_warnings=True,
  max_p=12, max_q=2, max_order=14,
  stepwise=False,
  seasonal=False
)

Code

ARIMA2 - Grid search all hyperparameters

• Default auto-arima may not be good to find best model
• Stepwise method may not be able to find the best model
• Conflict of accurate model and efficient model

Google Colaboratory