0.6.16 a59

TODO.md

@@ -28,8 +28,10 @@
 * 'round' option to Constraint
 * minor change to mosaic min style ensembles to remove edge case errors
 * 'mosaic-profile' and 'median' styles mosaic added
+* updated profiler, and improved feature generation for horizontal generalization
 * changepoint style trend as an option to GLM and GLS
 * added ShiftFirstValue which is only a minor nuance on PositiveShift transformer
+* added BasicLinearModel model
 
 ### Unstable Upstream Pacakges (those that are frequently broken by maintainers)
 * Pytorch-Forecasting

autots/models/basics.py

@@ -18,6 +18,7 @@
     create_changepoint_features,
     changepoint_fcst_from_last_row,
     random_datepart,
+    half_yr_spacing,
 )
 from autots.tools.probabilistic import Point_to_Probability, historic_quantile
 from autots.tools.window_functions import (
@@ -3434,11 +3435,11 @@ def fit(self, df, future_regressor=None):
         df = self.basic_profile(df)
         self.df = df
         if self.changepoint_spacing is None or self.changepoint_distance_end is None:
-            half_yr_spacing = int(df.shape[0] / ((df.index.max().year - df.index.min().year + 1) * 2))
+            half_yr_space = half_yr_spacing(df)
             if self.changepoint_spacing is None:
-                self.changepoint_spacing = half_yr_spacing
-            if self.changepoint_spacing is None:
-                self.changepoint_distance_end = int(half_yr_spacing / 2)
+                self.changepoint_spacing = int(half_yr_space)
+            if self.changepoint_distance_end is None:
+                self.changepoint_distance_end = int(half_yr_space / 2)
         if str(self.regression_type).lower() == "user":
             if future_regressor is None:
                 raise ValueError(
@@ -3457,7 +3458,7 @@ def fit(self, df, future_regressor=None):
         else:
             # Perform linear regression using the normal equation: (X.T @ X)^(-1) @ X.T @ Y
             self.beta = np.linalg.pinv(X_values.T @ X_values) @ X_values.T @ Y_values
-    
+
         # Calculate predicted values for Y
         Y_pred = X_values @ self.beta
 
@@ -3590,8 +3591,8 @@ def get_new_params(self, method: str = 'random'):
             )[0]
         return {
             "datepart_method": random_datepart(method=method),
-            "changepoint_spacing": random.choices([6, 28, 60, 90, 180, 360, 5040], [0.05, 0.1, 0.1, 0.1, 0.2, 0.1, 0.2])[0],
-            "changepoint_distance_end": random.choices([6, 28, 60, 90, 180, 360, 5040], [0.05, 0.1, 0.1, 0.1, 0.2, 0.1, 0.2])[0],
+            "changepoint_spacing": random.choices([None, 6, 28, 60, 90, 180, 360, 5040], [0.1, 0.05, 0.1, 0.1, 0.1, 0.2, 0.1, 0.2])[0],
+            "changepoint_distance_end": random.choices([None, 6, 28, 60, 90, 180, 360, 5040], [0.1, 0.05, 0.1, 0.1, 0.1, 0.2, 0.1, 0.2])[0],
             "regression_type": regression_choice,
             "λ": random.choices([None, 0.001, 0.01, 0.1, 1, 10, 100, 1000, 10000], [0.6, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1], k=1)[0],
         }

autots/tools/profile.py

@@ -4,11 +4,11 @@
 
 import numpy as np
 import pandas as pd
-from autots.tools.seasonal import date_part, create_changepoint_features
+from autots.tools.seasonal import date_part, create_changepoint_features, half_yr_spacing
 from autots.models.basics import BasicLinearModel
 
 def data_profile(df):
-    """
+    """Legacy profiler.
     Input: a pd DataFrame of columns which are time series, and a datetime index
 
     Output: a pd DataFrame of column per time series, with rows which are statistics
@@ -70,23 +70,35 @@ def summarize_series(df):
     return df_sum
 
 
-def profile_time_series(df, adi_threshold=1.3, cvar_threshold=0.5, flat_threshold=0.9, new_product_threshold=0.9, seasonal_threshold=0.5):
+def profile_time_series(
+        df, adi_threshold=1.3, cvar_threshold=0.5, flat_threshold=0.9,
+        new_product_threshold='auto',
+        seasonal_threshold=0.5
+):
     """
     Profiles time series data into categories: 
         smooth, intermittent, erratic, lumpy, flat, new_product
 
-    Parameters:
-    df (pd.DataFrame): Wide format DataFrame with datetime index and each column as a time series.
-
+    Args:
+        df (pd.DataFrame): Wide format DataFrame with datetime index and each column as a time series.
+        new_product_threshold (float): one of the more finiky thresholds, percent of null or zero data from beginning to declare new product
+        new_product_correct (bool): use dt index to correct
     Returns:
-    pd.DataFrame: DataFrame with 'SERIES' and 'DEMAND_PROFILE' columns.
+        pd.DataFrame: DataFrame with 'SERIES' and 'DEMAND_PROFILE' columns.
     """
 
     metrics_df = summarize_series(df).transpose()
 
     # Initialize demand profile as 'smooth'
     metrics_df['PROFILE'] = 'smooth'
 
+    if new_product_threshold == "auto":
+        half_yr_space = half_yr_spacing(df)
+        new_product_threshold = 1 - (half_yr_space * 0.65 / df.shape[0])
+        if new_product_threshold < 0.85:
+            new_product_threshold = 0.85
+        if new_product_threshold > 0.99:
+            new_product_threshold = 0.99
     # Apply conditions to classify the demand profiles
     metrics_df.loc[(metrics_df['adi'] >= adi_threshold) & (metrics_df['cv_squared'] < cvar_threshold), 'PROFILE'] = 'intermittent'
     metrics_df.loc[(metrics_df['adi'] < adi_threshold) & (metrics_df['cv_squared'] >= cvar_threshold), 'PROFILE'] = 'erratic'

autots/tools/seasonal.py

@@ -935,3 +935,6 @@ def changepoint_fcst_from_last_row(x_t_last_row, n_forecast=10):
     forecast_steps = np.arange(n_forecast).reshape(-1, 1)  # Shape it as multiple rows, 1 column
     extended_features = np.maximum(0, last_values + forecast_steps)
     return pd.DataFrame(extended_features, columns=x_t_last_row.index)
+
+def half_yr_spacing(df):
+    return int(df.shape[0] / ((df.index.max().year - df.index.min().year + 1) * 2))