dynabench.dataset.transforms

Functions

`abstractmethod`(funcobj)	A decorator indicating abstract methods.
`copy`(x)	Shallow copy operation on arbitrary Python objects.

Classes

`ABC`()	Helper class that provides a standard way to create an ABC using inheritance.
`BaseTransform`(**kwargs)
`CloudDataItem`(x[, y, pos, neighbors, ...])	Data class for 2D grid data.
`Compose`(transforms)	Compose function for combining multiple transforms.
`DataItem`(x[, y, pos])	Base class for data items.
`DefaultTransform`()	Default transformation for a data item.
`EdgeList`(k)	Create an edge list graph (src, dst) to use with PyG.
`EdgeListFromKNN`()	Create an edge list from the KNN graph.
`Grid2Cloud`()	Create a Cloud item from a grid data item
`GridDataItem`(x[, y, pos])	Data class for grid data.
`GridDownsampleFFT`([target_size])	Downsample the grid to a smaller size using FFT.
`GridDownsampleFactor`([factor])	Downsample the grid by a factor.
`Iterable`()
`KDTree`(data[, leafsize, compact_nodes, ...])	kd-tree for quick nearest-neighbor lookup.
`KNNGraph`(k[, grid_limits])	Create a KNN graph from the cloud data.
`PointSampling`([num_points])	Point sampling transform for the dataset.
`ToDict`()	Convert the data item to a dictionary.
`TypeCaster`(dtype)	Cast the data item to the correct type.

class dynabench.dataset.transforms.BaseTransform(**kwargs)[source]: Bases: ABC

class dynabench.dataset.transforms.Compose(transforms: List[BaseTransform])[source]

Bases: BaseTransform

Compose function for combining multiple transforms. Iterates over transformations and applies them to the data item.

Parameters:: transforms (List[BaseTransform]) – List of transforms to be applied to the data

class dynabench.dataset.transforms.DefaultTransform[source]

Bases: BaseTransform

Default transformation for a data item. Does not modify the data.

Parameters:: data_item (DataItem)
Returns:: transformed data_item
Return type:: DataItem

class dynabench.dataset.transforms.EdgeList(k: int)[source]

Bases: Compose

Create an edge list graph (src, dst) to use with PyG.

Parameters:: data_item (CloudItem)
Returns:: data_item with edge_list as knn_graph
Return type:: CloudItem

class dynabench.dataset.transforms.EdgeListFromKNN[source]

Bases: BaseTransform

Create an edge list from the KNN graph.

Parameters:: data_item (CloudItem)
Returns:: data_item with knn_graph
Return type:: CloudItem

check_if_valid()[source]

class dynabench.dataset.transforms.Grid2Cloud[source]

Bases: BaseTransform

Create a Cloud item from a grid data item

Parameters:: data_item (GridItem)
Returns:: data_item with cloud shape
Return type:: CloudItem

class dynabench.dataset.transforms.GridDownsampleFFT(target_size: Tuple[int, int] = (1.0, 1.0))[source]

Bases: BaseTransform

Downsample the grid to a smaller size using FFT.

Parameters:: target_size (Tuple[int, int]) – Target size of the grid.

class dynabench.dataset.transforms.GridDownsampleFactor(factor: int = 2)[source]

Bases: BaseTransform

Downsample the grid by a factor.

Parameters:: factor (int) – Factor by which to downsample the grid.

class dynabench.dataset.transforms.KNNGraph(k: int, grid_limits: Tuple[float] = (1.0, 1.0))[source]

Bases: BaseTransform

Create a KNN graph from the cloud data.

Parameters:: data_item (CloudItem)
Returns:: data_item with knn_graph
Return type:: CloudItem

check_if_valid()[source]

class dynabench.dataset.transforms.PointSampling(num_points: int = 900)[source]

Bases: BaseTransform

Point sampling transform for the dataset.

Parameters:

num_points (int) – Number of points to sample.
k (int) – Number of nearest neighbors to use for the KNN graph.

class dynabench.dataset.transforms.ToDict[source]

Bases: BaseTransform

Convert the data item to a dictionary.

Parameters:: data_item (DataItem)
Returns:: data_item as a dictionary
Return type:: dict

class dynabench.dataset.transforms.TypeCaster(dtype: ~numpy.dtype = <class 'numpy.float32'>)[source]

Bases: BaseTransform

Cast the data item to the correct type. (In place!!!)