CODEWITHSUNDEEP
arima
NewCoder1423
NewCoder1423

Reputation: 53

How to extract variables from the best auto_arima model to fit it?

I have a dataset with multiple cities and I'm trying to build an ARIMA model for each city, so in my code I'm splitting the data using a for loop and finding the best model before sending those parameters to the final fitting. My question is how can I automate the process? Is there any way to extract the p, d, q value out of the best model which is returned by the ARIMACheck function?

def ARIMACheck(data):
    from pmdarima import auto_arima  

    fit = auto_arima(data[20], trace=True)
    return fit

def ARIMA(data, p, d, q):    
    from statsmodels.tsa.arima.model import ARIMA
    
    x_train = data.iloc[:-200]
    x_test = data.iloc[-200:]
    y_test = x_test.loc[:, 20]
    model = ARIMA(x_train[20], order=(p,d,q))
    model = model.fit()
        
def Split(data):
    for i in range(7):
        data[i].replace(0, np.nan, inplace=True)
    
    for i in range(7):
        datatemp = data.copy()
        datatemp = datatemp.dropna(subset=[i])
        datamap = datatemp.copy()
        datamap = datamap.loc[:, 20]
        datamap.plot(figsize=(50,10))
        fit = ARIMACheck(datatemp)
        print(fit)
        ARIMA(datatemp, 1, 1, 2)

Split(data)

Upvotes: 2

Views: 3283

Answers (1)

Arne Decker
Arne Decker

Reputation: 923

First of all, the auto_arima function returns an ARIMA object that runs on statsmodels, so you could just use the fit from you method ARIMACheck(data).

If you want to create a new model with the statsmodels class, then you can use the following to extract the order from the auto_arima fit and use it to train a new model in your ARIMA method:

def ARIMA(data, fit):
    
    model = ARIMA(endog=x_train[20], order=fit.get_params().get("order")).fit()

and call the method by: ARIMA(datatemp, fit)

fit.get_params().get("order") returns a tuple like (p, d, q) so you can access each element by fit.get_params().get("order")[<index>].

Upvotes: 2

Related Questions