autogluon.timeseries 1.1.0b20240415__tar.gz → 1.1.0b20240417__tar.gz

{autogluon.timeseries-1.1.0b20240415 → autogluon.timeseries-1.1.0b20240417}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: autogluon.timeseries
-Version: 1.1.0b20240415
+Version: 1.1.0b20240417
 Summary: Fast and Accurate ML in 3 Lines of Code
 Home-page: https://github.com/autogluon/autogluon
 Author: AutoGluon Community

{autogluon.timeseries-1.1.0b20240415 → autogluon.timeseries-1.1.0b20240417}/src/autogluon/timeseries/configs/presets_configs.py

@@ -33,13 +33,13 @@ TIMESERIES_PRESETS_CONFIGS = dict(
     chronos_ensemble={
         "hyperparameters": {
             "Chronos": {"model_path": "small"},
-            **get_default_hps("default"),
+            **get_default_hps("light_inference"),
         }
     },
     chronos_large_ensemble={
         "hyperparameters": {
             "Chronos": {"model_path": "large", "batch_size": 8},
-            **get_default_hps("default"),
+            **get_default_hps("light_inference"),
         }
     },
 )

{autogluon.timeseries-1.1.0b20240415 → autogluon.timeseries-1.1.0b20240417}/src/autogluon/timeseries/models/autogluon_tabular/mlforecast.py

@@ -257,6 +257,10 @@ class AbstractMLForecastModel(AbstractTimeSeriesModel):
             if not df[col].isin([0, 1]).all():
                 df[f"__scaled_{col}"] = df[col] / df[col].abs().groupby(df[ITEMID]).mean().reindex(df[ITEMID]).values
 
+        # Convert float64 to float32 to reduce memory usage
+        float64_cols = list(df.select_dtypes(include="float64"))
+        df[float64_cols] = df[float64_cols].astype("float32")
+
         # We assume that df is sorted by 'unique_id' inside `TimeSeriesPredictor._check_and_prepare_data_frame`
         return df.rename(columns=column_name_mapping)
 

{autogluon.timeseries-1.1.0b20240415 → autogluon.timeseries-1.1.0b20240417}/src/autogluon/timeseries/models/local/abstract_local_model.py

@@ -230,8 +230,6 @@ def seasonal_naive_forecast(
         return arr[np.maximum.accumulate(idx)]
 
     forecast = {}
-    # Convert to float64 since std computation can be unstable in float32
-    target = target.astype(np.float64)
     # At least seasonal_period + 2 values are required to compute sigma for seasonal naive
     if len(target) > seasonal_period + 1 and seasonal_period > 1:
         if np.isnan(target[-(seasonal_period + 2) :]).any():

{autogluon.timeseries-1.1.0b20240415 → autogluon.timeseries-1.1.0b20240417}/src/autogluon/timeseries/models/presets.py

@@ -79,21 +79,23 @@ DEFAULT_MODEL_PRIORITY = dict(
     Average=100,
     SeasonalAverage=100,
     Zero=100,
-    NPTS=90,
-    ETS=90,
-    CrostonSBA=90,
-    Theta=80,
-    DynamicOptimizedTheta=80,
-    AutoETS=80,
-    AutoARIMA=70,
-    RecursiveTabular=60,
-    Chronos=50,
-    DirectTabular=50,
+    RecursiveTabular=90,
+    DirectTabular=85,
+    # All local models are grouped together to make sure that joblib parallel pool is reused
+    NPTS=80,
+    ETS=80,
+    CrostonSBA=80,
+    Theta=75,
+    DynamicOptimizedTheta=75,
+    AutoETS=70,
+    AutoARIMA=60,
+    Chronos=55,
+    # Models that can early stop are trained at the end
+    TemporalFusionTransformer=45,
     DeepAR=40,
-    TemporalFusionTransformer=30,
-    WaveNet=25,
-    PatchTST=20,
+    PatchTST=30,
     # Models below are not included in any presets
+    WaveNet=25,
     AutoCES=10,
     ARIMA=10,
     ADIDA=10,
@@ -128,6 +130,13 @@ def get_default_hps(key):
             "DirectTabular": {},
             "TemporalFusionTransformer": {},
         },
+        "light_inference": {
+            "SeasonalNaive": {},
+            "DirectTabular": {},
+            "RecursiveTabular": {},
+            "TemporalFusionTransformer": {},
+            "PatchTST": {},
+        },
         "default": {
             "SeasonalNaive": {},
             "CrostonSBA": {},
@@ -143,6 +152,7 @@ def get_default_hps(key):
             "TemporalFusionTransformer": {},
             "PatchTST": {},
             "DeepAR": {},
+            "Chronos": {"model_path": "base"},
         },
     }
     return default_model_hps[key]

{autogluon.timeseries-1.1.0b20240415 → autogluon.timeseries-1.1.0b20240417}/src/autogluon/timeseries/predictor.py

@@ -293,7 +293,7 @@ class TimeSeriesPredictor(TimeSeriesPredictorDeprecatedMixin):
             Preprocessed data in TimeSeriesDataFrame format.
         """
         df = self._to_data_frame(data, name=name)
-        df = df.astype({self.target: "float32"})
+        df = df.astype({self.target: "float64"})
         # MultiIndex.is_monotonic_increasing checks if index is sorted by ["item_id", "timestamp"]
         if not df.index.is_monotonic_increasing:
             df = df.sort_index()
@@ -499,7 +499,7 @@ class TimeSeriesPredictor(TimeSeriesPredictorDeprecatedMixin):
               with reasonable training time.
             - ``"high_quality"``: All ML models available in AutoGluon + additional statistical models (``NPTS``, ``AutoETS``, ``AutoARIMA``, ``CrostonSBA``,
               ``DynamicOptimizedTheta``). Much more accurate than ``medium_quality``, but takes longer to train.
-            - ``"best_quality"``: Same models as in ``"high_quality"`, but performs validation with multiple backtests. Usually better than ``high_quality``, but takes even longer to train.
+            - ``"best_quality"``: Same models as in ``"high_quality"``, but performs validation with multiple backtests. Usually better than ``high_quality``, but takes even longer to train.
 
             Available presets with the `Chronos <https://github.com/amazon-science/chronos-forecasting>`_ model:
 
@@ -507,8 +507,10 @@ class TimeSeriesPredictor(TimeSeriesPredictorDeprecatedMixin):
               See the documentation for ``ChronosModel`` or see `Hugging Face <https://huggingface.co/collections/amazon/chronos-models-65f1791d630a8d57cb718444>`_ for more information.
               Note that a GPU is required for model sizes ``small``, ``base`` and ``large``.
             - ``"chronos"``: alias for ``"chronos_small"``.
-            - ``"chronos_ensemble"``: builds an ensemble of the models specified in ``"high_quality"`` and ``"chronos_small"``.
-            - ``"chronos_large_ensemble"``: builds an ensemble of the models specified in ``"high_quality"`` and ``"chronos_large"``.
+            - ``"chronos_ensemble"``: builds an ensemble of seasonal naive, tree-based and deep learning models with fast inference
+              and ``"chronos_small"``.
+            - ``"chronos_large_ensemble"``: builds an ensemble of seasonal naive, tree-based and deep learning models
+              with fast inference and ``"chronos_large"``.
 
             Details for these presets can be found in ``autogluon/timeseries/configs/presets_configs.py``. If not
             provided, user-provided values for ``hyperparameters`` and ``hyperparameter_tune_kwargs`` will be used

{autogluon.timeseries-1.1.0b20240415 → autogluon.timeseries-1.1.0b20240417}/src/autogluon/timeseries/trainer/abstract_trainer.py

@@ -613,7 +613,7 @@ class AbstractTimeSeriesTrainer(SimpleAbstractTrainer):
                     break
 
             if random_seed is not None:
-                seed_everything(random_seed)
+                seed_everything(random_seed + i)
 
             if contains_searchspace(model.get_user_params()):
                 fit_log_message = f"Hyperparameter tuning model {model.name}. "

{autogluon.timeseries-1.1.0b20240415 → autogluon.timeseries-1.1.0b20240417}/src/autogluon/timeseries/utils/features.py

@@ -73,7 +73,7 @@ class ContinuousAndCategoricalFeatureGenerator(PipelineFeatureGenerator):
     Imputes missing categorical features with the most frequent value in the training set.
     """
 
-    def __init__(self, verbosity: int = 0, minimum_cat_count=2, float_dtype: str = "float32", **kwargs):
+    def __init__(self, verbosity: int = 0, minimum_cat_count=2, float_dtype: str = "float64", **kwargs):
         generators = [
             CategoryFeatureGenerator(minimum_cat_count=minimum_cat_count, fillna="mode"),
             IdentityFeatureGenerator(infer_features_in_args={"valid_raw_types": [R_INT, R_FLOAT]}),
@@ -111,7 +111,7 @@ class ContinuousAndCategoricalFeatureGenerator(PipelineFeatureGenerator):
 class TimeSeriesFeatureGenerator:
     """Takes care of preprocessing for static_features and past/known covariates.
 
-    All covariates & static features are converted into either float32 or categorical dtype.
+    All covariates & static features are converted into either float64 or categorical dtype.
 
     Missing values in the target column are left as-is but missing values in static features & covariates are imputed.
     Imputation logic is as follows:
@@ -121,16 +121,18 @@ class TimeSeriesFeatureGenerator:
         covariate values are missing, we fill them with the median of the training set.
     """
 
-    def __init__(self, target: str, known_covariates_names: List[str], float_dtype: str = "float32"):
+    def __init__(self, target: str, known_covariates_names: List[str], float_dtype: str = "float64"):
         self.target = target
         self.float_dtype = float_dtype
         self._is_fit = False
         self.known_covariates_names = list(known_covariates_names)
         self.past_covariates_names = []
-        self.known_covariates_pipeline = ContinuousAndCategoricalFeatureGenerator()
-        self.past_covariates_pipeline = ContinuousAndCategoricalFeatureGenerator()
+        self.known_covariates_pipeline = ContinuousAndCategoricalFeatureGenerator(float_dtype=float_dtype)
+        self.past_covariates_pipeline = ContinuousAndCategoricalFeatureGenerator(float_dtype=float_dtype)
         # Cat features with cat_count=1 are fine in static_features since they are repeated for all time steps in a TS
-        self.static_feature_pipeline = ContinuousAndCategoricalFeatureGenerator(minimum_cat_count=1)
+        self.static_feature_pipeline = ContinuousAndCategoricalFeatureGenerator(
+            minimum_cat_count=1, float_dtype=float_dtype
+        )
         self.covariate_metadata: CovariateMetadata = None
         self._train_covariates_real_median: Optional[pd.Series] = None
         self._train_static_real_median: Optional[pd.Series] = None

{autogluon.timeseries-1.1.0b20240415 → autogluon.timeseries-1.1.0b20240417}/src/autogluon/timeseries/version.py

@@ -1,3 +1,3 @@
 """This is the autogluon version file."""
-__version__ = '1.1.0b20240415'
+__version__ = '1.1.0b20240417'
 __lite__ = False

{autogluon.timeseries-1.1.0b20240415 → autogluon.timeseries-1.1.0b20240417}/src/autogluon.timeseries.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: autogluon.timeseries
-Version: 1.1.0b20240415
+Version: 1.1.0b20240417
 Summary: Fast and Accurate ML in 3 Lines of Code
 Home-page: https://github.com/autogluon/autogluon
 Author: AutoGluon Community

{autogluon.timeseries-1.1.0b20240415 → autogluon.timeseries-1.1.0b20240417}/src/autogluon.timeseries.egg-info/requires.txt

@@ -15,9 +15,9 @@ utilsforecast<0.0.11,>=0.0.10
 tqdm<5,>=4.38
 orjson~=3.9
 tensorboard<3,>=2.9
-autogluon.core[raytune]==1.1.0b20240415
-autogluon.common==1.1.0b20240415
-autogluon.tabular[catboost,lightgbm,xgboost]==1.1.0b20240415
+autogluon.core[raytune]==1.1.0b20240417
+autogluon.common==1.1.0b20240417
+autogluon.tabular[catboost,lightgbm,xgboost]==1.1.0b20240417
 
 [all]
 optimum[onnxruntime]<1.19,>=1.17