src.FRAME_FM.transforms.transforms

Attributes

DA
DS
TT
transform_mapping
apply_preprocessors

Classes

BaseTransform
FillMissingValueTransform
FillNaNTransform
NormalizeTransform
ScaleTransform
RenameTransform
ResampleTransform
ReshapeTransform
RollTransform
ReverseAxisTransform
SortAxisTransform
SubsetTransform
SqueezeTransform
TilerTransform	A transform that takes a Dataset or DataArray and breaks it into smaller tiles along specified dimensions.
ToDataArray
ToTensorTransform
TransposeTransform
VarsToDimensionTransform	A transform that takes a list of variables from a Dataset and stacks them into a

Functions

resolve_transform(→ BaseTransform)	If a transform is a dictionary with a "type" key, resolve it to the corresponding transform class instance.
apply_transforms(→ xarray.Dataset | xarray.DataArray)	Apply a list of preprocessing transforms to a data sample.

Module Contents

src.FRAME_FM.transforms.transforms.DA

src.FRAME_FM.transforms.transforms.DS

src.FRAME_FM.transforms.transforms.TT

class src.FRAME_FM.transforms.transforms.BaseTransform(*args, **kwargs)

abstractmethod __call__(sample)

class src.FRAME_FM.transforms.transforms.FillMissingValueTransform(strategy: str = 'constant', fill_value: None | float = None, method: None | str = 'linear')

Bases: BaseTransform

strategy = 'constant'

fill_value = None

method = 'linear'

__call__(sample: DS | DA) → DS | DA

class src.FRAME_FM.transforms.transforms.FillNaNTransform(strategy: str = 'constant', fill_value: None | float = None, method: None | str = 'linear')

Bases: FillMissingValueTransform

class src.FRAME_FM.transforms.transforms.NormalizeTransform(mean: float, std: float)

Bases: BaseTransform

mean

std

__call__(sample: DA) → DA

class src.FRAME_FM.transforms.transforms.ScaleTransform(mean: float, std: float)

Bases: NormalizeTransform

class src.FRAME_FM.transforms.transforms.RenameTransform(var_id: str, new_name: str)

Bases: BaseTransform

var_id

new_name

__call__(sample: DS) → DS

class src.FRAME_FM.transforms.transforms.ResampleTransform(dim: str, freq: str | int, method: str = 'mean')

Bases: BaseTransform

dim

freq

method = 'mean'

__call__(sample)

class src.FRAME_FM.transforms.transforms.ReshapeTransform(shape: tuple)

Bases: BaseTransform

shape

__call__(sample)

class src.FRAME_FM.transforms.transforms.RollTransform(dim: str, shift: None | int)

Bases: BaseTransform

dim

shift

__call__(sample)

class src.FRAME_FM.transforms.transforms.ReverseAxisTransform(dim: str)

Bases: BaseTransform

dim

__call__(sample)

class src.FRAME_FM.transforms.transforms.SortAxisTransform(dim: str, ascending: bool = True)

Bases: BaseTransform

dim

ascending = True

__call__(sample)

class src.FRAME_FM.transforms.transforms.SubsetTransform(**subset_selectors)

Bases: BaseTransform

subset_selectors

__call__(sample)

class src.FRAME_FM.transforms.transforms.SqueezeTransform(*args, **kwargs)

Bases: BaseTransform

__call__(sample)

class src.FRAME_FM.transforms.transforms.TilerTransform(boundary: str = 'pad', **dim_tile_sizes)

Bases: BaseTransform

A transform that takes a Dataset or DataArray and breaks it into smaller tiles along specified dimensions. This uses the xarray coarsen + construct pattern to create non-overlapping tiles of the data, which can be useful for training models on large spatial datasets by reducing memory usage and allowing for batch processing of smaller chunks of data.

boundary = 'pad'

tile_sizes

__call__(sample: DA) → DA

class src.FRAME_FM.transforms.transforms.ToDataArray(var_id: str)

Bases: BaseTransform

var_id

__call__(sample: DS | DA) → DA

class src.FRAME_FM.transforms.transforms.ToTensorTransform(*args, **kwargs)

Bases: BaseTransform

__call__(sample: DA | numpy.ndarray) → torch.Tensor

class src.FRAME_FM.transforms.transforms.TransposeTransform(*args, **kwargs)

Bases: BaseTransform

__call__(sample)

class src.FRAME_FM.transforms.transforms.VarsToDimensionTransform(variables: list, new_dim: str, only_vars_with_time: bool = True)

Bases: BaseTransform

A transform that takes a list of variables from a Dataset and stacks them into a new dimension, effectively converting the variable dimension into a coordinate dimension. This is useful for models that expect a single multi-channel input rather than separate variables.

Since the purpose is to prepare the data for conversion to a Tensor, we assume that ancillary variables that are not genuine coordinates can be dropped.

exclusion_vars = ['time_bounds', 'lat_bounds', 'lon_bounds', 'time_bnds', 'lat_bnds', 'lon_bnds', 'crs',...

variables

new_dim

only_vars_with_time = True

__call__(sample)

src.FRAME_FM.transforms.transforms.transform_mapping

src.FRAME_FM.transforms.transforms.resolve_transform(transform_config: dict) → BaseTransform

If a transform is a dictionary with a “type” key, resolve it to the corresponding transform class instance. If it is already an instance of a transform class, return it as is. Args: - transform_config (dict or BaseTransform): The transform configuration to resolve. Returns: - BaseTransform: An instance of a transform class.

src.FRAME_FM.transforms.transforms.apply_transforms(data: xarray.Dataset | xarray.DataArray, preprocessors: list) → xarray.Dataset | xarray.DataArray

Apply a list of preprocessing transforms to a data sample. :param sample: The input data sample to be transformed. :type sample: xr.Dataset | xr.DataArray :param preprocessors: A list of transform configurations to apply to the sample. :type preprocessors: list

Returns:

The transformed data sample after applying all preprocessors.

Return type:

xr.Dataset | xr.DataArray

src.FRAME_FM.transforms.transforms.apply_preprocessors