pyshred.engine package#

Submodules#

pyshred.engine.engine module#

class pyshred.engine.engine.SHREDEngine(data_manager: DataManager, shred_model: SHRED)[source]#

Bases: object

High-level interface for SHRED model inference and evaluation.

Parameters:

  • data_manager (DataManager) – Prepared data manager with fitted scalers.

  • shred_model (SHRED) – Trained SHRED model instance.

Variables:

  • dm (DataManager) – The data manager for preprocessing and postprocessing.

  • model (SHRED) – The trained SHRED model.

decode(latents)[source]#

Decode latent states back to full physical state space.

Parameters:

latents (np.ndarray or torch.Tensor) – Latent representations to decode.

Returns:

Dictionary mapping dataset IDs to reconstructed physical states.

Return type:

dict

evaluate(sensor_measurements: ndarray, Y: Dict[str, ndarray]) → DataFrame[source]#

Performs end‐to‐end reconstruction error in the physical space.

Parameters:

  • sensor_measurements ((T, n_sensors)) – The test sensor time series.

  • Y (dict[id] -> array (T, *spatial_shape)) – The raw full‐state ground truth for each dataset id.

Return type:

DataFrame indexed by dataset id with columns [MSE, RMSE, MAE, R2].

forecast_latent(h, init_latents)[source]#

Forecast future latent states using the latent forecaster.

Parameters:

  • h (int) – Number of future timesteps to forecast.

  • init_latents (np.ndarray or torch.Tensor) – Initial latent states for seeding the forecast.

Returns:

Forecasted latent states.

Return type:

np.ndarray

Raises:

RuntimeError – If no latent forecaster is available.

sensor_to_latent(sensor_measurements: ndarray | Tensor | DataFrame) → ndarray[source]#

Convert raw sensor measurements into latent-space embeddings.

Parameters:

sensor_measurements (array-like of shape (T, n_sensors)) – Raw sensor time series.

Returns:

latents – The GRU/LSTM final-hidden-state at each time index.

Return type:

np.ndarray of shape (T, latent_dim)

pyshred.engine.parametric_engine module#

class pyshred.engine.parametric_engine.ParametricSHREDEngine(data_manager: ParametricDataManager, shred_model: SHRED)[source]#

Bases: object

High-level interface for SHRED model inference and evaluation.

Parameters:

  • data_manager (ParametricDataManager) – Prepared data manager with fitted scalers.

  • shred_model (SHRED) – Trained SHRED model instance.

Variables:

  • dm (ParametricDataManager) – The data manager for preprocessing and postprocessing.

  • model (SHRED) – The trained SHRED model.

decode(latents)[source]#

Decode latent states back to full physical state space.

Parameters:

latents (np.ndarray or torch.Tensor) – Latent representations to decode.

Returns:

Dictionary mapping dataset IDs to reconstructed physical states.

Return type:

dict

evaluate(sensor_measurements: ndarray, Y: Dict[str, ndarray]) → DataFrame[source]#

Performs end‐to‐end reconstruction error in the physical space.

Parameters:

  • sensor_measurements ((T, n_sensors)) – The test sensor time series.

  • Y (dict[id] -> array (T, *spatial_shape)) – The raw full‐state ground truth for each dataset id.

Return type:

DataFrame indexed by dataset id with columns [MSE, RMSE, MAE, R2].

sensor_to_latent(sensor_measurements: ndarray | Tensor | DataFrame) → ndarray[source]#

Convert raw sensor measurements into latent-space embeddings.

Parameters:

sensor_measurements (array-like of shape (T, n_sensors)) – Raw sensor time series.

Returns:

latents – The GRU/LSTM final-hidden-state at each time index.

Return type:

np.ndarray of shape (T, latent_dim)

Module contents#