flash.core.data¶

flash.core.data.base_viz ¶

This Base Class is used to create visualization tool on top of InputTransform hooks.

This function is used to uncollate a batch into samples.

`BaseDataFetcher`	This class is used to profile `InputTransform` hook outputs.
`ControlFlow`
`FlashCallback`	`FlashCallback` is an extension of `pytorch_lightning.callbacks.Callback`.

A basic DataModule class for all Flash tasks.

`TargetFormatter`	A `TargetFormatter` is used to convert targets of a given type to a standard format required by the loss function.
`SingleNumericTargetFormatter`	A `TargetFormatter` for targets that contain a single numeric value (the class index).
`SingleLabelTargetFormatter`	A `TargetFormatter` for targets that contain a single string label.
`SingleBinaryTargetFormatter`	A `TargetFormatter` for targets that are one-hot encoded binaries.
`MultiNumericTargetFormatter`	A `TargetFormatter` for targets that contain multiple numeric values (the class indices).
`MultiLabelTargetFormatter`	A `TargetFormatter` for targets that contain multiple string labels in a list.
`CommaDelimitedMultiLabelTargetFormatter`	A `TargetFormatter` for targets that contain a string with multiple comma-delimited labels.
`SpaceDelimitedTargetFormatter`	A `TargetFormatter` for targets that contain a string with multiple space-delimited labels.
`MultiBinaryTargetFormatter`	A `TargetFormatter` for targets that are multi-hot binary.
`MultiSoftTargetFormatter`	A `TargetFormatter` for mutli-label soft targets.
`get_target_formatter`	Get the `TargetFormatter` object to use for the given targets.

`wrap_collate`	`flash.data.utilities.collate.wrap_collate()` is a utility that can be used to wrap an existing collate function to handle the metadata separately from the rest of the batch (giving a list of the metadata from the samples in the output).
`default_collate`	The `flash.data.utilities.collate.default_collate()` extends torch.utils.data._utils.default_collate to first extract any metadata from the samples in the batch (in the `"metadata"` key).

`load_image`	Load an image from a file.
`load_spectrogram`	Load a spectrogram from an image or audio file.
`load_audio`	Load a waveform from an audio file.
`load_data_frame`	Load a data frame from a CSV (or similar) file.

SplitDataset is used to create Dataset Subset using indices.

The ApplyToKeys class is an nn.Sequential which applies the given transforms to the given keys from the input.

This class is used to wrap a callable within a nn.Module and apply the wrapped function in __call__

`convert_to_modules`
`download_data`	Download file with progressbar.

`Input`
`DataKeys`	The `DataKeys` enum contains the keys that are used by built-in data sources to refer to inputs and targets.
`InputFormat`	The `InputFormat` enum contains the data source names used by all of the default `from_*` methods in `DataModule`.

The ClassificationInputMixin class provides utility methods for handling classification targets.

An Output encapsulates a single transform() method which is used to convert the model output into the desired output format when predicting.

The OutputTransform encapsulates all the data processing logic that should run after the model.