`amisc.typing`

Module with type hints for the AMISC package.

Includes:

MultiIndex — tuples of integers or similar string representations
Dataset — a type hint for the input/output dicts handled by Component.model
TrainIteration — the results of a single training iteration
CompressionData — a dictionary spec for passing data to/from Variable.compress()
LATENT_STR_ID — a string identifier for latent coefficients of field quantities

`CompressionData`

Bases: TypedDict

Configuration dict for passing compression data to/from Variable.compress().

Field quantity shapes

Field quantity data can take on any arbitrary shape, which we indicate with qty.shape. For example, a 3d structured grid might have qty.shape = (10, 15, 10). Unstructured data might just have qty.shape = (N,) for \(N\) points in an unstructured grid. Regardless, Variable.compress() will flatten this and compress to a single latent vector of size latent_size. That is, qty.shape → latent_size.

Compression coordinates

Field quantity data must be specified along with its coordinate locations. If the coordinate locations are different from what was used when building the compression map (i.e. the SVD data matrix), then they will be interpolated to/from the SVD coordinates.

ATTRIBUTE	DESCRIPTION
`coords`	`(qty.shape, dim)` the coordinate locations of the qty data; coordinates exist in `dim` space (e.g. `dim=2` for 2d Cartesian coordinates). Defaults to the coordinates used when building the construction map (i.e. the coordinates of the data in the SVD data matrix) TYPE: `ndarray`
`latent`	`(..., latent_size)` array of latent space coefficients for a field quantity; this is what is returned by `Variable.compress()` and what is expected as input by `Variable.reconstruct()`. TYPE: `ndarray`
`qty`	`(..., qty.shape)` array of uncompressed field quantity data for this qty within the `fields` list. Each qty in this list will be its own `key:value` pair in the `CompressionData` structure TYPE: `ndarray`

`Dataset`

Bases: TypedDict

Type hint for the input/output dicts of a call to Component.model. The keys are the variable names and the values are the corresponding np.ndarrays. There are also a few special keys that can be returned by the model that are described below.

The model can return additional items that are not part of Component.outputs. These items are returned as object arrays in the output.

This data structure is very similar to the Dataset class in the xarray package. Later versions might consider migrating to xarray for more advanced data manipulation.

ATTRIBUTE	DESCRIPTION
`model_cost`	the computational cost (seconds of CPU time) of a single model evaluation TYPE: `float \| list \| ArrayLike`
`output_path`	the path to the output file or directory written by the model TYPE: `str \| Path`
`errors`	a `dict` with the indices where the model evaluation failed with context about the errors TYPE: `dict`

`MultiIndex`

Bases: tuple

A multi-index is a tuple of integers, can be converted from a string.

`TrainIteration`

Bases: TypedDict

Gives the results of a single training iteration.

ATTRIBUTE	DESCRIPTION
`component`	the name of the component selected for refinement at this iteration TYPE: `str`
`alpha`	the selected candidate model fidelity multi-index TYPE: `MultiIndex`
`beta`	the selected candidate surrogate fidelity multi-index TYPE: `MultiIndex`
`num_evals`	the number of model evaluations performed during this iteration TYPE: `int`
`added_cost`	the total added computational cost of the new model evaluations TYPE: `float`
`added_error`	the error/difference between the refined surrogate and the previous surrogate TYPE: `float`
`test_error`	the error of the refined surrogate on the test set (optional) TYPE: `Optional[dict[str, float]]`