deepblink.util module

Utility helper functions.

deepblink.util.delete_non_unique_columns(df: pandas.core.frame.DataFrame) → pandas.core.frame.DataFrame

Deletes DataFrame columns that only contain one (non-unique) value.

deepblink.util.get_from_module(path: str, attribute: str) → Callable

Grab an attribute (e.g. class) from a given module path.

deepblink.util.predict_pixel_size(fname: Union[str, os.PathLike[str]]) → Tuple[float, float]

Predict the pixel size based on tifffile metadata.

deepblink.util.predict_shape(shape: tuple) → str

Predict the channel-arangement based on common standards.

Assumes the following things: * x, y are the two largest axes * rgb only if the last axis is 3 * up to 4 channels * “fill up order” is c, z, t

Parameters:shape – To be predicted shape. Output from np.ndarray.shape

deepblink.util.relative_shuffle(x: Union[list, numpy.ndarray], y: Union[list, numpy.ndarray]) → Tuple[Union[list, numpy.ndarray], Union[list, numpy.ndarray]]

Shuffles x and y keeping their relative order.

deepblink.util.remove_falses(tup: tuple) → tuple

Removes all false occurences from a tuple.

deepblink.util.train_valid_split(x_list: list, y_list: list, valid_split: float = 0.2, shuffle: bool = True) → Iterable[list]

Split two lists (usually input and ground truth).

Splitting into random training and validation sets with an optional shuffling.

Parameters:

• x_list – First list of items. Typically input data.
• y_list – Second list of items. Typically labeled data.
• valid_split – Number between 0-1 to denote the percentage of examples used for validation.

Returns:

(x_train, x_valid, y_train, y_valid) splited lists containing training or validation examples respectively.