autogluon.timeseries 1.4.1b20250917__py3-none-any.whl → 1.4.1b20251003__py3-none-any.whl

autogluon/timeseries/models/toto/model.py

@@ -0,0 +1,234 @@
+import logging
+import os
+from typing import TYPE_CHECKING, Any, Optional, Sequence, Union
+
+import numpy as np
+import pandas as pd
+from typing_extensions import Self
+
+from autogluon.common.loaders import load_pkl
+from autogluon.timeseries import TimeSeriesDataFrame
+from autogluon.timeseries.models.abstract import AbstractTimeSeriesModel
+from autogluon.timeseries.utils.features import CovariateMetadata
+
+if TYPE_CHECKING:
+    from ._internal import TotoForecaster
+
+logger = logging.getLogger(__name__)
+
+
+class TotoModel(AbstractTimeSeriesModel):
+    """Toto (Time-Series-Optimized Transformer for Observability) [CohenKhwajaetal2025]_ pretrained time series forecasting model.
+
+    Toto is a 151M parameter model trained on over 1T data points from DataDog's internal observability systems, as well as
+    the GIFT-eval pretrain, Chronos pretraining, and synthetically generated time series corpora. It is a decoder-only
+    architecture that autoregressively outputs parametric distribution forecasts. More details can be found on
+    `Hugging Face <https://huggingface.co/Datadog/Toto-Open-Base-1.0>`_ and `GitHub <https://github.com/DataDog/toto>`_.
+
+    The AutoGluon implementation of Toto is on a port of the original implementation. It is optimized for easy maintenance
+    with the rest of the AutoGluon model zoo, and does not feature some important optimizations such as xformers and flash-attention
+    available in the original model repository. The AutoGluon implementation of Toto requires a CUDA-compatible GPU.
+
+    References
+    ----------
+    .. [CohenKhwajaetal2025] Cohen, Ben, Khwaja, Emaad et al.
+        "This Time is Different: An Observability Perspective on Time Series Foundation Models."
+        https://arxiv.org/abs/2505.14766
+
+
+    Other Parameters
+    ----------------
+    model_path : str, default = "Datadog/Toto-Open-Base-1.0"
+        Model path used for the model, i.e., a HuggingFace transformers ``name_or_path``. Can be a
+        compatible model name on HuggingFace Hub or a local path to a model directory.
+    batch_size : int, default = 24
+        Size of batches used during inference.
+    num_samples : int, default = 256
+        Number of samples used during inference.
+    device : str, default = "cuda"
+        Device to use for inference. Toto requires a CUDA-compatible GPU to run.
+    context_length : int or None, default = 4096
+        The context length to use in the model. Shorter context lengths will decrease model accuracy, but result
+        in faster inference.
+    compile_model : bool, default = True
+        Whether to compile the model using torch.compile() for faster inference. May increase initial loading time
+        but can provide speedups during inference.
+    """
+
+    default_model_path: str = "Datadog/Toto-Open-Base-1.0"
+
+    def __init__(
+        self,
+        path: Optional[str] = None,
+        name: Optional[str] = None,
+        hyperparameters: Optional[dict[str, Any]] = None,
+        freq: Optional[str] = None,
+        prediction_length: int = 1,
+        covariate_metadata: Optional[CovariateMetadata] = None,
+        target: str = "target",
+        quantile_levels: Sequence[float] = (0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9),
+        eval_metric: Any = None,
+    ):
+        hyperparameters = hyperparameters if hyperparameters is not None else {}
+
+        self.model_path = hyperparameters.get("model_path", self.default_model_path)
+
+        super().__init__(
+            path=path,
+            name=name,
+            hyperparameters=hyperparameters,
+            freq=freq,
+            prediction_length=prediction_length,
+            covariate_metadata=covariate_metadata,
+            target=target,
+            quantile_levels=quantile_levels,
+            eval_metric=eval_metric,
+        )
+
+        self._forecaster: Optional[TotoForecaster] = None
+
+    def save(self, path: Optional[str] = None, verbose: bool = True) -> str:
+        forecaster = self._forecaster
+        self._forecaster = None
+        path = super().save(path=path, verbose=verbose)
+        self._forecaster = forecaster
+
+        return str(path)
+
+    @classmethod
+    def load(cls, path: str, reset_paths: bool = True, load_oof: bool = False, verbose: bool = True) -> Self:
+        model = load_pkl.load(path=os.path.join(path, cls.model_file_name), verbose=verbose)
+        if reset_paths:
+            model.set_contexts(path)
+
+        return model
+
+    def _is_gpu_available(self) -> bool:
+        import torch.cuda
+
+        return torch.cuda.is_available()
+
+    def get_minimum_resources(self, is_gpu_available: bool = False) -> dict[str, Union[int, float]]:
+        return {"num_cpus": 1, "num_gpus": 1}
+
+    def load_forecaster(self):
+        from ._internal import TotoForecaster
+        from .hf_pretrained_model import TotoConfig, TotoPretrainedModel
+
+        if not self._is_gpu_available():
+            raise RuntimeError(
+                f"{self.name} requires a GPU to run, but no GPU was detected. "
+                "Please make sure that you are using a computer with a CUDA-compatible GPU and "
+                "`import torch; torch.cuda.is_available()` returns `True`."
+            )
+
+        hyperparameters = self.get_hyperparameters()
+        pretrained_model = TotoPretrainedModel.from_pretrained(
+            self.model_path,
+            config=TotoConfig.from_pretrained(self.model_path),
+            device_map=hyperparameters["device"],
+        )
+
+        if hyperparameters["compile_model"]:
+            pretrained_model.model.compile()
+
+        self._forecaster = TotoForecaster(model=pretrained_model.model)
+
+    def persist(self) -> Self:
+        if self._forecaster is None:
+            self.load_forecaster()
+        return self
+
+    def _get_default_hyperparameters(self) -> dict:
+        return {
+            "batch_size": 24,
+            "num_samples": 256,
+            "device": "cuda",
+            "context_length": 4096,
+            "compile_model": True,
+        }
+
+    @property
+    def allowed_hyperparameters(self) -> list[str]:
+        return super().allowed_hyperparameters + [
+            "model_path",
+            "batch_size",
+            "num_samples",
+            "device",
+            "context_length",
+            "compile_model",
+        ]
+
+    def _more_tags(self) -> dict:
+        return {
+            "allow_nan": True,
+            "can_use_train_data": False,
+            "can_use_val_data": False,
+        }
+
+    def _fit(
+        self,
+        train_data: TimeSeriesDataFrame,
+        val_data: Optional[TimeSeriesDataFrame] = None,
+        time_limit: Optional[float] = None,
+        num_cpus: Optional[int] = None,
+        num_gpus: Optional[int] = None,
+        verbosity: int = 2,
+        **kwargs,
+    ) -> None:
+        self._check_fit_params()
+        self.load_forecaster()
+
+    def _predict(
+        self, data: TimeSeriesDataFrame, known_covariates: Optional[TimeSeriesDataFrame] = None, **kwargs
+    ) -> TimeSeriesDataFrame:
+        import torch
+
+        from .dataloader import TotoDataLoader, TotoInferenceDataset
+
+        hyperparameters = self.get_hyperparameters()
+
+        if self._forecaster is None:
+            self.load_forecaster()
+        assert self._forecaster, "Toto model failed to load"
+        device = self._forecaster.model.device
+
+        dataset = TotoInferenceDataset(
+            target_df=data.fill_missing_values("auto"),
+            max_context_length=hyperparameters["context_length"],
+        )
+        loader = TotoDataLoader(
+            dataset,
+            freq=self.freq,
+            batch_size=hyperparameters["batch_size"],
+            time_limit=kwargs.get("time_limit"),
+            device=device,
+        )
+
+        batch_means, batch_quantiles = [], []
+        with torch.inference_mode():
+            for masked_timeseries in loader:
+                forecast = self._forecaster.forecast(
+                    masked_timeseries,
+                    prediction_length=self.prediction_length,
+                    num_samples=hyperparameters["num_samples"],
+                    samples_per_batch=32,
+                )
+
+                batch_means.append(forecast.mean.cpu().numpy())
+                qs = np.array([forecast.quantile(q).cpu().numpy() for q in self.quantile_levels])
+                batch_quantiles.append(qs.squeeze(2).transpose(1, 2, 0))
+
+        df = pd.DataFrame(
+            np.concatenate(
+                [
+                    np.concatenate(batch_means, axis=0).reshape(-1, 1),
+                    np.concatenate(batch_quantiles, axis=0).reshape(-1, len(self.quantile_levels)),
+                ],
+                axis=1,
+            ),
+            columns=["mean"] + [str(q) for q in self.quantile_levels],
+            index=self.get_forecast_horizon_index(data),
+        )
+
+        return TimeSeriesDataFrame(df)

autogluon/timeseries/version.py

@@ -1,4 +1,4 @@
 """This is the autogluon version file."""
 
-__version__ = "1.4.1b20250917"
+__version__ = "1.4.1b20251003"
 __lite__ = False

{autogluon.timeseries-1.4.1b20250917.dist-info → autogluon.timeseries-1.4.1b20251003.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: autogluon.timeseries
-Version: 1.4.1b20250917
+Version: 1.4.1b20251003
 Summary: Fast and Accurate ML in 3 Lines of Code
 Home-page: https://github.com/autogluon/autogluon
 Author: AutoGluon Community
@@ -55,16 +55,21 @@ Requires-Dist: fugue>=0.9.0
 Requires-Dist: tqdm<5,>=4.38
 Requires-Dist: orjson~=3.9
 Requires-Dist: tensorboard<3,>=2.9
-Requires-Dist: autogluon.core[raytune]==1.4.1b20250917
-Requires-Dist: autogluon.common==1.4.1b20250917
-Requires-Dist: autogluon.features==1.4.1b20250917
-Requires-Dist: autogluon.tabular[catboost,lightgbm,xgboost]==1.4.1b20250917
+Requires-Dist: autogluon.core[raytune]==1.4.1b20251003
+Requires-Dist: autogluon.common==1.4.1b20251003
+Requires-Dist: autogluon.features==1.4.1b20251003
+Requires-Dist: autogluon.tabular[catboost,lightgbm,xgboost]==1.4.1b20251003
 Provides-Extra: all
+Requires-Dist: rotary-embedding-torch<1,>=0.8; extra == "all"
+Requires-Dist: einops<1,>=0.7; extra == "all"
 Provides-Extra: tests
 Requires-Dist: pytest; extra == "tests"
 Requires-Dist: ruff>=0.0.285; extra == "tests"
 Requires-Dist: flaky<4,>=3.7; extra == "tests"
 Requires-Dist: pytest-timeout<3,>=2.1; extra == "tests"
+Provides-Extra: toto
+Requires-Dist: einops<1,>=0.7; extra == "toto"
+Requires-Dist: rotary-embedding-torch<1,>=0.8; extra == "toto"
 
 
 

{autogluon.timeseries-1.4.1b20250917.dist-info → autogluon.timeseries-1.4.1b20251003.dist-info}/RECORD

@@ -1,11 +1,11 @@
-autogluon.timeseries-1.4.1b20250917-py3.9-nspkg.pth,sha256=cQGwpuGPqg1GXscIwt-7PmME1OnSpD-7ixkikJ31WAY,554
+autogluon.timeseries-1.4.1b20251003-py3.9-nspkg.pth,sha256=cQGwpuGPqg1GXscIwt-7PmME1OnSpD-7ixkikJ31WAY,554
 autogluon/timeseries/__init__.py,sha256=_CrLLc1fkjen7UzWoO0Os8WZoHOgvZbHKy46I8v_4k4,304
 autogluon/timeseries/evaluator.py,sha256=l642tYfTHsl8WVIq_vV6qhgAFVFr9UuZD7gLra3A_Kc,250
 autogluon/timeseries/learner.py,sha256=eQrqFVOmL-2JC85LgCMkbyoLpKS02Dilg1T8RUeS_LI,13887
 autogluon/timeseries/predictor.py,sha256=7X4YsWYa3Xk2RI1Irf2O-c3-I82Zqhg-cgj8cj_4AoA,88427
 autogluon/timeseries/regressor.py,sha256=lc8Qr3-8v4oxajtCnV3sxpUaW6vxXXJOA6Kr-qVne4k,11926
 autogluon/timeseries/splitter.py,sha256=8ACkuCXeUhQGUx4jz_Vv17q814WrHJQeKvq2v4-oE6s,3158
-autogluon/timeseries/version.py,sha256=4ybwz-_gJJkK6joDld6cMM3LMRqq9eYTxT7WGP327uM,91
+autogluon/timeseries/version.py,sha256=m7vaKYpgPZNMLb9FSMXLkDiT72Ij1eEgM8hr-FGDUbc,91
 autogluon/timeseries/configs/__init__.py,sha256=wiLBwxZkDTQBJkSJ9-xz3p_yJxX0dbHe108dS1P5O6A,183
 autogluon/timeseries/configs/hyperparameter_presets.py,sha256=GbI2sd3uakWtaeaMyF7B5z_lmyfb6ToK6PZEUZTyG9w,2031
 autogluon/timeseries/configs/predictor_presets.py,sha256=B5HFHIelh91hhG0YYE5SJ7_14P7sylFAABgHX8n_53M,2712
@@ -16,7 +16,7 @@ autogluon/timeseries/metrics/abstract.py,sha256=6jbluvHXfLc_cuK1Fx0ZYle2sR4WGG6Y
 autogluon/timeseries/metrics/point.py,sha256=sS__n_Em7m4CUaBu3PNWQ_dHw1YCOHbEyC15fhytFL8,18308
 autogluon/timeseries/metrics/quantile.py,sha256=x0cq44fXRoMiuI4BVQ7mpWk1YgrK4OwLTlJAhCHQ7Xg,4634
 autogluon/timeseries/metrics/utils.py,sha256=HuDe1BNe8yJU4f_DKM913nNrUueoRaw6zhxm1-S20s0,910
-autogluon/timeseries/models/__init__.py,sha256=9YnqkOILtVEkbICk7J3VlMkMNySs-f5ErIUKrE5-fys,1294
+autogluon/timeseries/models/__init__.py,sha256=9NY9mqYaZe_7XB70M6psHARH-Lpkfroj4toUUPO9BmI,1339
 autogluon/timeseries/models/registry.py,sha256=8n7W04ql0ckNQUzKcAW7bxreLI8wTAUTymACgLklH9M,2158
 autogluon/timeseries/models/abstract/__init__.py,sha256=Htfkjjc3vo92RvyM8rIlQ0PLWt3jcrCKZES07UvCMV0,146
 autogluon/timeseries/models/abstract/abstract_timeseries_model.py,sha256=97HOi7fRPxtx8Y9hq-xdJI-kLMp6Z-8LUSvcfBjXFsM,31978
@@ -49,6 +49,21 @@ autogluon/timeseries/models/local/npts.py,sha256=VRZk5tEJOIentt0tLM6lxyoU8US736n
 autogluon/timeseries/models/local/statsforecast.py,sha256=sZ6aEFzAyPNZX3rMULGWFht0Toapjb3EwHe5Rb76ZxA,33318
 autogluon/timeseries/models/multi_window/__init__.py,sha256=Bq7AT2Jxdd4WNqmjTdzeqgNiwn1NCyWp4tBIWaM-zfI,60
 autogluon/timeseries/models/multi_window/multi_window_model.py,sha256=Hn-H2jLdeuB0_TxhAdununS8ti-iO-WSl3FOoxzcEJA,12369
+autogluon/timeseries/models/toto/__init__.py,sha256=rQaVjZJV5ZsJGC0jhQ6CA4nYeXdV1KtlyDz2i2usQnY,54
+autogluon/timeseries/models/toto/dataloader.py,sha256=A5WHhnAe0J7fPo2KKG43hYLSrtUBGNweuqxMmClu3_A,3598
+autogluon/timeseries/models/toto/hf_pretrained_model.py,sha256=Q8bVUaSlQVE4xFn_v7H0h_NFTxzHiM1V17KFytc50jk,4783
+autogluon/timeseries/models/toto/model.py,sha256=eP0SAoUjv9l_ExK4eoPl9ZZHW_MXa-OVLYxhj3f1bl4,8809
+autogluon/timeseries/models/toto/_internal/__init__.py,sha256=tKkiux9bD2Xu0AuVyTEx_sNOZutcluC7-d7tn7wsmec,193
+autogluon/timeseries/models/toto/_internal/dataset.py,sha256=xuAEOhoQNJGMoCxkLVLrgpdoOJuukAYbrSrnrkwFob0,6103
+autogluon/timeseries/models/toto/_internal/forecaster.py,sha256=UXiohiySn_Gs8kLheeVcVCO8qoEtYlEfMH1tukAOHsk,18520
+autogluon/timeseries/models/toto/_internal/backbone/__init__.py,sha256=hq5W62boH6HiEP8z3sHkI6_KM-Dd6TkDfWDm6DYE3J8,63
+autogluon/timeseries/models/toto/_internal/backbone/attention.py,sha256=HLUFoyqR8EqxUMT1BK-AjI4ClS8au35LcUo7Jx7Xhm0,9394
+autogluon/timeseries/models/toto/_internal/backbone/backbone.py,sha256=HUjpY2ZWed74UYKjp31erXF2ZHf3mmQMw_5_cCFeJGg,10104
+autogluon/timeseries/models/toto/_internal/backbone/distribution.py,sha256=8NXiaEVLuvjTW7L1t1RzooZFNERWv50zyLddbAwuYpo,2502
+autogluon/timeseries/models/toto/_internal/backbone/kvcache.py,sha256=QSVCrnbS2oD7wkJodZbP9XMVmrfCH6M3Zp44siF28Fg,5399
+autogluon/timeseries/models/toto/_internal/backbone/rope.py,sha256=Ghngo08DjHbwbyp6b-GXCyLeYR10dH-Y_RMOTYwIxPY,3527
+autogluon/timeseries/models/toto/_internal/backbone/scaler.py,sha256=opqyhHIZ6mPdPlrr3gA0qt9FFogIAYNDSq-P7CyQiqE,13728
+autogluon/timeseries/models/toto/_internal/backbone/transformer.py,sha256=5c-ngj4XHKlaedz1NkgdfQgqD2kUGkMn4mtGH_lTXsE,12410
 autogluon/timeseries/trainer/__init__.py,sha256=_tw3iioJfvtIV7wnjtEMv0yS8oabmCFxDnGRodYE7RI,72
 autogluon/timeseries/trainer/model_set_builder.py,sha256=s6tozfND3lLfst6Vxa_oP_wgCmDapyCJYFmCjkEn-es,10788
 autogluon/timeseries/trainer/prediction_cache.py,sha256=Vi6EbMiMheq_smA93U_MoMxYUV85RdPm0dvJFdsM8K4,5551
@@ -65,11 +80,11 @@ autogluon/timeseries/utils/datetime/base.py,sha256=3NdsH3NDq4cVAOSoy3XpaNixyNlbj
 autogluon/timeseries/utils/datetime/lags.py,sha256=rjJtdBU0M41R1jwfmvCbo045s-6XBjhGVnGBQJ9-U1E,5997
 autogluon/timeseries/utils/datetime/seasonality.py,sha256=YK_2k8hvYIMW-sJPnjGWRtCnvIOthwA2hATB3nwVoD4,834
 autogluon/timeseries/utils/datetime/time_features.py,sha256=kEOFls4Nzh8nO0Pcz1DwLsC_NA3hMI4JUlZI3kuvuts,2666
-autogluon.timeseries-1.4.1b20250917.dist-info/LICENSE,sha256=CeipvOyAZxBGUsFoaFqwkx54aPnIKEtm9a5u2uXxEws,10142
-autogluon.timeseries-1.4.1b20250917.dist-info/METADATA,sha256=_55BDBngt0x8tNYPNYFDOT_hn5iSUXCsn0BrJN4F9_s,12463
-autogluon.timeseries-1.4.1b20250917.dist-info/NOTICE,sha256=7nPQuj8Kp-uXsU0S5so3-2dNU5EctS5hDXvvzzehd7E,114
-autogluon.timeseries-1.4.1b20250917.dist-info/WHEEL,sha256=tZoeGjtWxWRfdplE7E3d45VPlLNQnvbKiYnx7gwAy8A,92
-autogluon.timeseries-1.4.1b20250917.dist-info/namespace_packages.txt,sha256=giERA4R78OkJf2ijn5slgjURlhRPzfLr7waIcGkzYAo,10
-autogluon.timeseries-1.4.1b20250917.dist-info/top_level.txt,sha256=giERA4R78OkJf2ijn5slgjURlhRPzfLr7waIcGkzYAo,10
-autogluon.timeseries-1.4.1b20250917.dist-info/zip-safe,sha256=AbpHGcgLb-kRsJGnwFEktk7uzpZOCcBY74-YBdrKVGs,1
-autogluon.timeseries-1.4.1b20250917.dist-info/RECORD,,
+autogluon.timeseries-1.4.1b20251003.dist-info/LICENSE,sha256=CeipvOyAZxBGUsFoaFqwkx54aPnIKEtm9a5u2uXxEws,10142
+autogluon.timeseries-1.4.1b20251003.dist-info/METADATA,sha256=xAIDouWR1x06xs7OinjyC1cQwqxSt2IZDeEWyC-slKw,12702
+autogluon.timeseries-1.4.1b20251003.dist-info/NOTICE,sha256=7nPQuj8Kp-uXsU0S5so3-2dNU5EctS5hDXvvzzehd7E,114
+autogluon.timeseries-1.4.1b20251003.dist-info/WHEEL,sha256=tZoeGjtWxWRfdplE7E3d45VPlLNQnvbKiYnx7gwAy8A,92
+autogluon.timeseries-1.4.1b20251003.dist-info/namespace_packages.txt,sha256=giERA4R78OkJf2ijn5slgjURlhRPzfLr7waIcGkzYAo,10
+autogluon.timeseries-1.4.1b20251003.dist-info/top_level.txt,sha256=giERA4R78OkJf2ijn5slgjURlhRPzfLr7waIcGkzYAo,10
+autogluon.timeseries-1.4.1b20251003.dist-info/zip-safe,sha256=AbpHGcgLb-kRsJGnwFEktk7uzpZOCcBY74-YBdrKVGs,1
+autogluon.timeseries-1.4.1b20251003.dist-info/RECORD,,