Time-Series Models#

SimbaML provides a variety of models for time-series prediction. These models can be used for the synthetic-data pipeline and mixed-data pipeline.

Note

Time-Series models in SimbaML are models that are trained once on either synthetic or a mix of synthetic and real-world data. In constrast, transfer learning time-series models are first trained on synthetic and then fine-tuned on real-world data. Here, the model’s train() function is called twice.

Provided Models for Time-Series Prediction#

The following models can be used out-of-the-box for the synthetic-data pipeline and mixed-data pipeline. Note that the strings indicate the model’s id that is used in the config file to select the model for training.

  • Scikit-learn

    • Linear Regressor: “LinearRegressor”

    • Random Forest: “RandomForestRegressor”

    • Decision Tree: “DecisionTreeRegressor”

    • Support Vector Machine: “SVMRegressor”

    • Nearest Neighbor: “NearestNeighborRegressor”

  • Pytorch Lightning

    • Deep Neural Network: “PytorchLightningDenseNeuralNetwork”

  • Keras

    • Deep Neural Network: “KerasDenseNeuralNetwork”

Note

To successfully run the Pytorch Lightning and Keras deep neural network, you have to install the corresponding packages first. Both are not installed by default when installing SimbaML.

Besides, we provide two benchmark models:

  • Last Value Predictor: “LastValuePredictor”

  • Average Value Predictor “AverageValuePredictor”

Include Own Model for Time-Series Prediction#

SimbaML allows for the effortless integration of any other machine learning models for time-series prediction. This includes PyTorch Lightning, Keras, and Scikit-learn models. The following code show how to integrate a model, which always predicts zero:

>>> import dataclasses
>>> import numpy as np
>>> import numpy.typing as npt
>>>
>>> from simba_ml.prediction.time_series.models import model
>>> from simba_ml.prediction.time_series.models import factory
>>>
>>>
>>> @dataclasses.dataclass
... class ZeroPredictorConfig(model.ModelConfig):
...     """Defines the configuration for the DenseNeuralNetwork."""
...     name: str = "Zero Predictor"
...
...
>>> class ZeroPredictor(model.Model):
...     """Defines a model, which predicts the average of the train data."""
...
...     def __init__(self, input_length: int, output_length: int, config: ZeroPredictorConfig):
...         """Inits the `AveragePredictor`.
...
...         Args:
...             input_length: the length of the input data.
...             output_length: the length of the output data.
...             config: the config for the model
...         """
...         super().__init__(input_length, output_length, config)
...
...     def set_seed(self, seed: int) -> None:
...         """Sets the seed for the model. For this model, this is not required."""
...         pass
...
...     def train(self, train: list[npt.NDArray[np.float64]], val: list[npt.NDArray[np.float64]]) -> None:
...         pass
...
...     def predict(self, data: npt.NDArray[np.float64]) -> npt.NDArray[np.float64]:
...         self.validate_prediction_input(data)
...         return np.full((data.shape[0], self.output_length, data.shape[2]), 0.0)
...
...
>>> def register() -> None:
...     factory.register(
...         "ZeroPredictor",
...         ZeroPredictorConfig,
...         ZeroPredictor
...     )