Update for package with fixes

forecastout/config.yaml

@@ -6,7 +6,7 @@ models:
   autoarima:
     "seasonality": "True"
     "freq": 12
-    "alpha_intervals": 0.05
+    "alpha_intervals": 95
   prophet:
   holtwinters:
     "freq": 12

forecastout/core.py

@@ -150,8 +150,8 @@ def __get_daily_disaggregation_shares(self):
             df=df.copy(),
             current_month=(
                 self.df_monthly.loc[
-                    ~self.df_monthly['value'].isna(), 'date'
-                ].max()
+                    self.df_monthly['value'].isna(), 'date'
+                ].min()
             ),
             granularity_month=False
         )
@@ -192,5 +192,6 @@ def __make_daily_forecast(self):
     def __remake_monthly_forecast(self):
         self.df_monthly_forecast = remake_monthly_forecast_current_month(
             df_daily_forecast=self.df_daily_forecast.copy(),
-            df_actuals=self.df.copy()
+            df_actuals=self.df.copy(),
+            sum_aggregation=self.sum_aggregation
         )

forecastout/disaggregation_engine/remake_monthly_forecast_current_month.py

@@ -3,7 +3,8 @@
 
 def remake_monthly_forecast_current_month(
         df_daily_forecast: pd.DataFrame,
-        df_actuals: pd.DataFrame
+        df_actuals: pd.DataFrame,
+        sum_aggregation: bool
 ) -> pd.DataFrame:
     # -- Get actuals of unfinished month
     first_forecast_day = df_daily_forecast['date'].min()
@@ -26,13 +27,22 @@ def remake_monthly_forecast_current_month(
     df_daily_forecast["date"] = (
         df_daily_forecast["date"].astype(str).str[0:7]
     )
-    df_monthly_forecast = (
-        df_daily_forecast
-        .groupby(["date", "model"])
-        [["forecast", "forecast_lower", "forecast_upper"]]
-        .sum()
-        .reset_index()
-    )
+    if sum_aggregation:
+        df_monthly_forecast = (
+            df_daily_forecast
+            .groupby(["date", "model"])
+            [["forecast", "forecast_lower", "forecast_upper"]]
+            .sum()
+            .reset_index()
+        )
+    else:
+        df_monthly_forecast = (
+            df_daily_forecast
+            .groupby(["date", "model"])
+            [["forecast", "forecast_lower", "forecast_upper"]]
+            .mean()
+            .reset_index()
+        )
     df_monthly_forecast['date'] = pd.to_datetime(
         df_monthly_forecast['date'] + '-01'
     )

forecastout/forecast_engine/forecast_model_factory.py

@@ -19,7 +19,8 @@ def get_model(
         if model == "autoarima":
             return AutoArimaModel(
                 df_train_y=df_train_y,
-                dict_config=dict_config
+                dict_config=dict_config,
+                series_train_dates=series_train_dates
             )
         if model == "prophet":
             return ProphetModel(

forecastout/forecast_models/abstract_model.py

@@ -2,15 +2,18 @@
 from abc import ABC, abstractmethod
 import pandas as pd
 
+
 class ForecastModel(ABC):
 
     def __init__(
             self,
-            df_train_y:pd.DataFrame,
-            dict_config:dict
+            df_train_y: pd.DataFrame,
+            dict_config: dict,
+            series_train_dates: pd.Series
     ):
         self.df_train_y = df_train_y
         self.dict_config = dict_config
+        self.series_train_dates = series_train_dates
 
     @abstractmethod
     def do_forecast(self, list_dates: list):

forecastout/forecast_models/autoarima_model.py

@@ -1,7 +1,7 @@
 from forecastout.forecast_models.abstract_model import ForecastModel
-from pmdarima.arima import auto_arima
+from statsforecast.models import AutoARIMA
+from statsforecast import StatsForecast
 import pandas as pd
-import numpy as np
 
 
 class AutoArimaModel(ForecastModel):
@@ -13,34 +13,50 @@ def __init__(self,
                  **kwargs
                  ):
         super(AutoArimaModel, self).__init__(*args, **kwargs)
-        self.arima_model = (
-            auto_arima(self.df_train_y,
-                       seasonal=self.dict_config["seasonality"],
-                       m=self.dict_config["freq"]
-                       )
+        # -- Constructor
+        # -- Initialize autoarima
+        self.autoarima_model = StatsForecast(models=[
+            AutoARIMA(
+                season_length=self.dict_config['freq'],
+                seasonal=self.dict_config['seasonality'])],
+            freq='MS',
+            n_jobs=-1
         )
+        # -- Transform data
+        df_train_y_autoarima = (
+            pd.concat([self.series_train_dates, self.df_train_y], axis=1)
+        )
+        df_train_y_autoarima.columns = ['ds', 'y']
+        df_train_y_autoarima['ds'] = (
+            pd.to_datetime(df_train_y_autoarima['ds'])
+        )
+        df_train_y_autoarima['ds'] = (
+            df_train_y_autoarima['ds'].dt.tz_localize(None)
+        )
+        df_train_y_autoarima['unique_id'] = 'id'
+        # -- Fit forecast_models
+        self.autoarima_model.fit(df_train_y_autoarima)
 
     def do_forecast(self, list_dates: list) -> pd.DataFrame:
-        arima_prediction, arima_conf_int = self.arima_model.predict(
-            len(list_dates),
-            return_conf_int=True,
-            alpha=self.dict_config["alpha_intervals"]
+        # -- Predict
+        autoarima_prediction = self.autoarima_model.predict(
+            h=len(list_dates),
+            level=[self.dict_config['alpha_intervals']]
+        ).reset_index()
+        # -- Transform data for correct output
+        autoarima_prediction = (
+            autoarima_prediction.drop(['unique_id'], axis=1)
         )
-        arima_prediction = pd.concat(
-            [
-                pd.DataFrame(
-                    np.array(arima_prediction.reset_index()[0]),
-                    columns=['forecast']),
-                pd.DataFrame(
-                    arima_conf_int,
-                    columns=['forecast_lower', 'forecast_upper'])
-            ],
-            axis=1)
-        arima_prediction['model'] = 'autoarima'
-        arima_prediction['date'] = list_dates
+        autoarima_prediction.columns = [
+            'date', 'forecast', 'forecast_lower', 'forecast_upper'
+        ]
+        autoarima_prediction['model'] = 'autoarima'
+        autoarima_prediction['date'] = list_dates
         # -- Ensure date
-        arima_prediction['date'] = pd.to_datetime(arima_prediction['date'])
-        return arima_prediction
+        autoarima_prediction['date'] = (
+            pd.to_datetime(autoarima_prediction['date'])
+        )
+        return autoarima_prediction
 
     def get_feature_importance(self):
         return None

forecastout/forecast_models/holtwinters_model.py

@@ -7,7 +7,6 @@
 
 class HoltWintersModel(ForecastModel):
     def __init__(self,
-                 series_train_dates: pd.Series,
                  *args,
                  **kwargs
                  ):
@@ -33,7 +32,7 @@ def __init__(self,
         )
         # -- Transform data
         df_train_y_holtwinters = pd.concat(
-            [series_train_dates, self.df_train_y],
+            [self.series_train_dates, self.df_train_y],
             axis=1
         )
         df_train_y_holtwinters.columns = ['ds', 'y']

forecastout/forecast_models/naive_seasonal.py

@@ -6,7 +6,6 @@
 
 class NaiveSeasonalModel(ForecastModel):
     def __init__(self,
-                 series_train_dates: pd.Series,
                  *args,
                  **kwargs
                  ):
@@ -24,7 +23,7 @@ def __init__(self,
         )
         # -- Transform data
         df_train_y_seasonalnaive = (
-            pd.concat([series_train_dates, self.df_train_y], axis=1)
+            pd.concat([self.series_train_dates, self.df_train_y], axis=1)
         )
         df_train_y_seasonalnaive.columns = ['ds', 'y']
         df_train_y_seasonalnaive['ds'] = (

forecastout/forecast_models/prophet_model.py

@@ -5,7 +5,6 @@
 
 class ProphetModel(ForecastModel):
     def __init__(self,
-                 series_train_dates: pd.Series,
                  *args,
                  **kwargs
                  ):
@@ -16,7 +15,7 @@ def __init__(self,
         # -- Prophet
         # -- Constructor
         df_train_y_prophet = pd.concat(
-            [series_train_dates, self.df_train_y],
+            [self.series_train_dates, self.df_train_y],
             axis=1)
         df_train_y_prophet.columns = ['ds', 'y']
         df_train_y_prophet['ds'] = pd.to_datetime(df_train_y_prophet['ds'])

tests/disaggregation_engine/test_remake_monthly_forecast_current_month.py

@@ -8,7 +8,8 @@ def test_remake_monthly_forecast_current_month():
     pd.testing.assert_frame_equal(
         remake_monthly_forecast_current_month(
             df_daily_forecast=input_df_daily_forecast(),
-            df_actuals=input_df_actuals()
+            df_actuals=input_df_actuals(),
+            sum_aggregation=True
         ),
         expected_df()
     )