tfds.features.FeatureConnector

Abstract base class for feature types.

tfds.features.FeatureConnector(
    *, doc: DocArg = None
)

This class provides an interface between the way the information is stored on disk, and the way it is presented to the user.

Here is a diagram on how FeatureConnector methods fit into the data generation/reading:

generator => encode_example() => tf_example => decode_example() => data dict

The connector can either get raw or dictionary values as input, depending on the connector type.

doc

dtype Return the dtype (or dict of dtype) of this FeatureConnector.
np_dtype

numpy_dtype

shape Return the shape (or dict of shape) of this FeatureConnector.
tf_dtype

Methods

catalog_documentation

View source

catalog_documentation() -> List[CatalogFeatureDocumentation]

Returns the feature documentation to be shown in the catalog.

cls_from_name

View source

@classmethod
cls_from_name(
    python_class_name: str
) -> Type['FeatureConnector']

Returns the feature class for the given Python class.

decode_batch_example

View source

decode_batch_example(
    tfexample_data
)

Decode multiple features batched in a single tf.Tensor.

This function is used to decode features wrapped in tfds.features.Sequence(). By default, this function apply decode_example on each individual elements using tf.map_fn. However, for optimization, features can overwrite this method to apply a custom batch decoding.

Args
tfexample_data Same tf.Tensor inputs as decode_example, but with and additional first dimension for the sequence length.

Returns
tensor_data Tensor or dictionary of tensor, output of the tf.data.Dataset object

decode_example

View source

decode_example(
    tfexample_data
)

Decode the feature dict to TF compatible input.

Args
tfexample_data Data or dictionary of data, as read by the tf-example reader. It correspond to the tf.Tensor() (or dict of tf.Tensor()) extracted from the tf.train.Example, matching the info defined in get_serialized_info().

Returns
tensor_data Tensor or dictionary of tensor, output of the tf.data.Dataset object

decode_example_np

View source

decode_example_np(
    example_data: type_utils.NpArrayOrScalar
) -> type_utils.NpArrayOrScalar

Encode the feature dict into NumPy-compatible input.

Args
example_data Value to convert to NumPy.

Returns
np_data Data as NumPy-compatible type: either a Python primitive (bytes, int, etc) or a NumPy array.

decode_ragged_example

View source

decode_ragged_example(
    tfexample_data
)

Decode nested features from a tf.RaggedTensor.

This function is used to decode features wrapped in nested tfds.features.Sequence(). By default, this function apply decode_batch_example on the flat values of the ragged tensor. For optimization, features can overwrite this method to apply a custom batch decoding.

Args
tfexample_data tf.RaggedTensor inputs containing the nested encoded examples.

Returns
tensor_data The decoded tf.RaggedTensor or dictionary of tensor, output of the tf.data.Dataset object

encode_example

View source

@abc.abstractmethod
encode_example(
    example_data
)

Encode the feature dict into tf-example compatible input.

The input example_data can be anything that the user passed at data generation. For example:

For features:

features={
    'image': tfds.features.Image(),
    'custom_feature': tfds.features.CustomFeature(),
}

At data generation (in _generate_examples), if the user yields:

yield {
    'image': 'path/to/img.png',
    'custom_feature': [123, 'str', lambda x: x+1]
}

Then

Args
example_data Value or dictionary of values to convert into tf-example compatible data.

Returns
tfexample_data Data or dictionary of data to write as tf-example. Data can be a list or numpy array. Note that numpy arrays are flattened so it's the feature connector responsibility to reshape them in decode_example(). Note that tf.train.Example only supports int64, float32 and string so the data returned here should be integer, float or string. User type can be restored in decode_example().

from_config

View source

@classmethod
from_config(
    root_dir: str
) -> FeatureConnector

Reconstructs the FeatureConnector from the config file.

Usage:

features = FeatureConnector.from_config('path/to/dir')

Args
root_dir Directory containing the features.json file.

Returns
The reconstructed feature instance.

from_json

View source

@classmethod
from_json(
    value: Json
) -> FeatureConnector

FeatureConnector factory.

This function should be called from the tfds.features.FeatureConnector base class. Subclass should implement the from_json_content.

Example:

feature = tfds.features.FeatureConnector.from_json(
    {'type': 'Image', 'content': {'shape': [32, 32, 3], 'dtype': 'uint8'} }
)
assert isinstance(feature, tfds.features.Image)

Args
value dict(type=, content=) containing the feature to restore. Match dict returned by to_json.

Returns
The reconstructed FeatureConnector.

from_json_content

View source

@classmethod
from_json_content(
    value: Union[Json, message.Message], doc: Optional[DocArg] = None
) -> T

FeatureConnector factory (to overwrite).

Subclasses should overwrite this method. This method is used when importing the feature connector from the config.

This function should not be called directly. FeatureConnector.from_json should be called instead.

See existing FeatureConnectors for implementation examples.

Args
value FeatureConnector information represented as either Json or a Feature proto. The content must match what is returned by to_json_content.
doc Documentation of this feature (e.g. description).

Returns
The reconstructed FeatureConnector.

from_proto

View source

@classmethod
from_proto(
    feature_proto: feature_pb2.Feature
) -> T

Instantiates a feature from its proto representation.

get_serialized_info

View source

get_serialized_info() -> Union[TensorInfo, Mapping[str, TensorInfo]]

Return the shape/dtype of features after encoding (for the adapter).

The FileAdapter then use those information to write data on disk.

This function indicates how this feature is encoded on file internally. The DatasetBuilder are written on disk as tf.train.Example proto.

Ex:

return {
    'image': tfds.features.TensorInfo(shape=(None,), dtype=np.uint8),
    'height': tfds.features.TensorInfo(shape=(), dtype=np.int32),
    'width': tfds.features.TensorInfo(shape=(), dtype=np.int32),
}

FeatureConnector which are not containers should return the feature proto directly:

return tfds.features.TensorInfo(shape=(64, 64), np.uint8)

If not defined, the retuned values are automatically deduced from the get_tensor_info function.

Returns
features Either a dict of feature proto object, or a feature proto object

get_tensor_info

View source

@abc.abstractmethod
get_tensor_info() -> TreeDict[TensorInfo]

Return the tf.Tensor dtype/shape of the feature.

This returns the tensor dtype/shape, as returned by .as_dataset by the tf.data.Dataset object.

Ex:

return {
    'image': tfds.features.TensorInfo(shape=(None,), dtype=np.uint8),
    'height': tfds.features.TensorInfo(shape=(), dtype=np.int32),
    'width': tfds.features.TensorInfo(shape=(), dtype=np.int32),
}

FeatureConnector which are not containers should return the feature proto directly:

return tfds.features.TensorInfo(shape=(256, 256), dtype=np.uint8)

Returns
tensor_info Either a dict of tfds.features.TensorInfo object, or a tfds.features.TensorInfo

get_tensor_spec

View source

get_tensor_spec() -> TreeDict[tf.TensorSpec]

Returns the tf.TensorSpec of this feature (not the element spec!).

Note that the output of this method may not correspond to the element spec of the dataset. For example, currently this method does not support RaggedTensorSpec.

load_metadata

View source

load_metadata(
    data_dir: epath.PathLike, feature_name: Optional[str]
)

Restore the feature metadata from disk.

If a dataset is re-loaded and generated files exists on disk, this function will restore the feature metadata from the saved file.

Args
data_dir path to the dataset folder to which save the info (ex: ~/datasets/cifar10/1.2.0/)
feature_name the name of the feature (from the FeaturesDict key)

repr_html

View source

repr_html(
    ex: np.ndarray
) -> str

Returns the HTML str representation of the object.

repr_html_batch

View source

repr_html_batch(
    ex: np.ndarray
) -> str

Returns the HTML str representation of the object (Sequence).

repr_html_ragged

View source

repr_html_ragged(
    ex: np.ndarray
) -> str

Returns the HTML str representation of the object (Nested sequence).

save_config

View source

save_config(
    root_dir: str
) -> None

Exports the FeatureConnector to a file.

Args
root_dir path/to/dir containing the features.json

save_metadata

View source

save_metadata(
    data_dir: epath.PathLike, feature_name: Optional[str]
) -> None

Save the feature metadata on disk.

This function is called after the data has been generated (by _download_and_prepare) to save the feature connector info with the generated dataset.

Some dataset/features dynamically compute info during _download_and_prepare. For instance:

  • Labels are loaded from the downloaded data
  • Vocabulary is created from the downloaded data
  • ImageLabelFolder compute the image dtypes/shape from the manual_dir

After the info have been added to the feature, this function allow to save those additional info to be restored the next time the data is loaded.

By default, this function do not save anything, but sub-classes can overwrite the function.

Args
data_dir path to the dataset folder to which save the info (ex: ~/datasets/cifar10/1.2.0/)
feature_name the name of the feature (from the FeaturesDict key)

to_json

View source

to_json() -> Json

Exports the FeatureConnector to Json.

Each feature is serialized as a dict(type=..., content=...).

  • type: The cannonical name of the feature (module.FeatureName).
  • content: is specific to each feature connector and defined in to_json_content. Can contain nested sub-features (like for tfds.features.FeaturesDict and tfds.features.Sequence).

For example:

tfds.features.FeaturesDict({
    'input': tfds.features.Image(),
    'target': tfds.features.ClassLabel(num_classes=10),
})

Is serialized as:

{
    "type": "tensorflow_datasets.core.features.features_dict.FeaturesDict",
    "content": {
        "input": {
            "type": "tensorflow_datasets.core.features.image_feature.Image",
            "content": {
                "shape": [null, null, 3],
                "dtype": "uint8",
                "encoding_format": "png"
            }
        },
        "target": {
            "type":
            "tensorflow_datasets.core.features.class_label_feature.ClassLabel",
            "content": {
              "num_classes": 10
            }
        }
    }
}

Returns
A dict(type=, content=). Will be forwarded to from_json when reconstructing the feature.

to_json_content

View source

to_json_content() -> Union[Json, message.Message]

FeatureConnector factory (to overwrite).

This function should be overwritten by the subclass to allow re-importing the feature connector from the config. See existing FeatureConnector for example of implementation.

Returns
The FeatureConnector metadata in either a dict, or a Feature proto. This output is used in from_json_content when reconstructing the feature.

to_proto

View source

to_proto() -> feature_pb2.Feature

Exports the FeatureConnector to the Feature proto.

For features that have a specific schema defined in a proto, this function needs to be overriden. If there's no specific proto schema, then the feature will be represented using JSON.

Returns
The feature proto describing this feature.

ALIASES []