<forecast_context>
<dataset>
- Observations: 120
- Observations pooled across series: 60000
- Series: 500
- Frequency: D
- Date range: 2023-01-01 to 2023-04-30
- Target: ['series_000', 'series_001', 'series_002', 'series_003', 'series_004', 'series_005', 'series_006', 'series_007', 'series_008', 'series_009', 'series_010', 'series_011', 'series_012', 'series_013', 'series_014'] (first 15 of 500)
- Exogenous columns: exog_00, exog_01, exog_02, exog_03, exog_04, exog_05, exog_06, exog_07, exog_08, exog_09, exog_10, exog_11, exog_12, exog_13, exog_14 (first 15 of 20)
- Target statistics (series_000): min 74.05, max 123.3, mean 101.1, std 11.35
- Target statistics (series_001): min 64.41, max 127, mean 99.32, std 10.7
- Target statistics (series_002): min 78.57, max 121, mean 98.99, std 9.194
- Target statistics (series_003): min 74, max 124.7, mean 101.7, std 10.6
- Target statistics (series_004): min 70.73, max 127, mean 99.71, std 10.53
- Target statistics shown (first 5 of 500 series)
- Missing in target: {'series_000': 1, 'series_001': 1, 'series_002': 1, 'series_003': 1, 'series_004': 1, 'series_005': 1, 'series_006': 1, 'series_007': 1, 'series_008': 1, 'series_009': 1, 'series_010': 1, 'series_011': 1, 'series_012': 1, 'series_013': 1, 'series_014': 1} (first 15 of 30 series, 30 missing values in all)
- Index irregularities: none detected
</dataset>
<profile_decision>
The dataset contains 500 series, so a multi-series forecaster (ForecasterRecursiveMultiSeries) is recommended. Data: 60000 observations pooled across 500 series, 'D' frequency. Alternative forecasters: ['ForecasterDirectMultiVariate', 'ForecasterFoundation']. Estimator: LGBMRegressor. A gradient boosting model is preferred for a dataset of this size (60000 observations). Alternative estimators: ['XGBRegressor', 'Ridge']. 20 exogenous variables available as predictors.
- Significant lags shown (first 5 of 500 series; a series without significant lags has no line)
- Suggested window features: mean(window=3), std(window=3), mean(window=7), mean(window=21)
- Suggested calendar features: day_of_week, weekend, month
</profile_decision>
<forecast_plan>
- Forecaster: ForecasterRecursiveMultiSeries
- Steps: 3
- Estimator: LGBMRegressor
- Lags: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15] (first 15 of 30)
- Window features: [{'stats': ['mean', 'std'], 'window_size': 3}, {'stats': ['mean'], 'window_size': 7}, {'stats': ['mean'], 'window_size': 21}]
- Primary metric: mean_absolute_scaled_error
- Preprocessing steps:
  - [informational] Impute or handle missing values before training. Use dropna_from_series=True or a NaN-tolerant estimator.
- Chosen by the user instead of the rules: lags
- Plan: ForecasterRecursiveMultiSeries + LGBMRegressor. Lags: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30]. Window features: ['mean(window=3)', 'std(window=3)', 'mean(window=7)', 'mean(window=21)']. Calendar features: ['day_of_week', 'weekend', 'month'] (raw ordinal encoding). NaN rows kept (NaN-tolerant estimator). Exogenous variables included. MASE is scale-independent, enabling fair comparison across differently-scaled series.
</forecast_plan>
<backtesting_strategy>
- steps: 5
- initial_train_size: 70
- refit: False
- fixed_train_size: True
- gap: 0
- n_folds: 6
</backtesting_strategy>
<deterministic_summary>
Backtested with 6 folds of 5 steps each, starting from an initial training window of 70 observations, without refitting.
</deterministic_summary>
<evaluation_metrics>
Rows shown (first 5 of 500 series, plus the aggregated rows).
          levels  mean_absolute_error
      series_000                 5.00
      series_001                 5.01
      series_002                 5.02
      series_003                 5.03
      series_004                 5.04
         average                 7.50
weighted_average                 7.50
         pooling                 7.40
</evaluation_metrics>
<predictions>
Shape: 6 rows x 3 columns
Columns: ['level', 'fold', 'pred']
Folds: 1
  pred: min=97.5, max=102.5, mean=100.0, std=1.8708286933869707
Per-level summary of pred (all rows):
  series_000: min=100.5, max=102.5, mean=101.5
  series_001: min=97.5, max=99.5, mean=98.5
Index range: 2023-04-21 00:00:00 to 2023-04-23 00:00:00
</predictions>
</forecast_context>
