Reference

Top level¶

Every symbol below is importable directly from the resdag namespace. Each is a re-export; the canonical page listed alongside it documents it in full.

import resdag as rd

model = rd.ott_esn(reservoir_size=500, feedback_size=3, output_size=3)
trainer = rd.ESNTrainer(model)

Composition¶

Symbol	Canonical home
`ESN`	`resdag.facade`
`ESNModel`	`resdag.core`
`ReservoirFeatureExtractor`	`resdag.core`
`Input`	`resdag.core` (re-export of `pytorch_symbolic.Input`)
`reservoir_input`	`resdag.core`

Reservoirs and cells¶

Symbol	Canonical home
`ESNLayer`	`resdag.layers.reservoirs`
`NGReservoir`	`resdag.layers.reservoirs`
`BaseReservoirLayer`	`resdag.layers.reservoirs`
`ESNCell`	`resdag.layers.cells`
`NGCell`	`resdag.layers.cells`
`ReservoirCell`	`resdag.layers.cells`

Readouts and transforms¶

Symbol	Canonical home
`ReadoutLayer`	`resdag.layers.readouts`
`RidgeReadoutLayer`	`resdag.layers.readouts`
`CholeskyReadoutLayer`	`resdag.layers.readouts`
`CGReadoutLayer`	`resdag.layers.readouts`
`SVDReadoutLayer`	`resdag.layers.readouts`
`PinvReadoutLayer`	`resdag.layers.readouts`
`IncrementalRidgeReadout`	`resdag.layers.readouts`
`Concatenate`	`resdag.layers.transforms`
`FeaturePartitioner`	`resdag.layers.transforms`
`Power`	`resdag.layers.transforms`
`SelectiveDropout`	`resdag.layers.transforms`
`SelectiveExponentiation`	`resdag.layers.transforms`
`Standardize`	`resdag.layers.transforms`

Models and ensembles¶

Symbol	Canonical home
`classic_esn`	`resdag.models`
`ott_esn`	`resdag.models`
`power_augmented`	`resdag.models`
`linear_esn`	`resdag.models`
`headless_esn`	`resdag.models`
`coupled_ensemble_esn`	`resdag.models`
`CoupledEnsembleESNModel`	`resdag.ensemble`
`OutliersFilteredMean`	`resdag.ensemble.aggregators`

Training¶

Symbol	Canonical home
`ESNTrainer`	`resdag.training`

Data, datasets, and diagnostics¶

Symbol	Canonical home
`TimeSeriesWindowDataset`	`resdag.data`
`make_dataloader`	`resdag.data`
`lorenz`, `rossler`, `henon`, `mackey_glass`, `narma`, `sine`	`resdag.datasets`
`esp_index`	`resdag.utils.states`

The generators in resdag.datasets each return a (1, n_timesteps, features) tensor; their first positional argument is n_timesteps. TimeSeriesWindowDataset and make_dataloader are the torch DataLoader path — see Work · Streaming & DataLoaders.

Lazy HPO attributes¶

run_hpo, LOSSES, and get_study_summary are also reachable as rd.run_hpo, rd.LOSSES, and rd.get_study_summary. They are resolved lazily via module __getattr__, so optuna remains an optional dependency until one of these attributes is first accessed.

Submodules¶

resdag.core, resdag.layers, resdag.init, resdag.training, resdag.metrics, resdag.models, resdag.ensemble, resdag.hpo, and resdag.utils are all importable as attributes, alongside the convenience aliases resdag.graphs, resdag.topology, and resdag.input_feedback (re-exports of the matching resdag.init subpackages).