|
|
|
|
|
import pandas as pd |
|
|
from statsmodels.tsa.arima.model import ARIMAResults |
|
|
from sklearn.metrics import mean_absolute_percentage_error |
|
|
|
|
|
def load_model(path="best_arima_model.pkl"): |
|
|
return ARIMAResults.load(path) |
|
|
|
|
|
def load_timeseries(df, comm_code): |
|
|
monthly_cols = [col for col in df.columns if col.startswith('INDX')] |
|
|
df_filtered = df[df['COMM_CODE'] == comm_code] |
|
|
ts_values = df_filtered[monthly_cols].values.flatten() |
|
|
ts = pd.Series(ts_values, index=pd.date_range(start='2012-04-01', periods=len(monthly_cols), freq='MS')) |
|
|
return ts |
|
|
|
|
|
def forecast(comm_code, forecast_months=6, csv_path="data.csv"): |
|
|
df = pd.read_csv(csv_path) |
|
|
ts = load_timeseries(df, comm_code) |
|
|
train = ts[:-forecast_months] |
|
|
test = ts[-forecast_months:] |
|
|
|
|
|
model = load_model() |
|
|
forecast = model.forecast(steps=forecast_months) |
|
|
forecast.index = test.index |
|
|
|
|
|
mape = mean_absolute_percentage_error(test, forecast) |
|
|
return forecast, test, mape |
|
|
|
