deepblink.datasets package

Submodules

• deepblink.datasets.sequence module
• deepblink.datasets.spots module

Module contents

Datasets module with classes to handle data import and data presentation for training.

class deepblink.datasets.Dataset(name: str, *_)

Bases: object

Simple abstract class for datasets.

Parameters:name – Absolute path to dataset file.

data_filename

Return the absolute path to the dataset.

load_data()

Empty method to import or create data.

normalize_dataset()

Empty method to normalise images in the dataset.

prepare_data()

Empty method to prepare or convert data.

class deepblink.datasets.SequenceDataset(x: numpy.ndarray, y: numpy.ndarray, batch_size: int = 16, augment_fn: Callable = None, format_fn: Callable = None, overfit: bool = False)

Bases: keras.utils.data_utils.Sequence

Custom Sequence class used to feed data into model.fit.

Parameters:

• x_list – List of inputs.
• y_list – List of targets.
• batch_size – Size of one mini-batch.
• augment_fn – Function to augment one mini-batch of x and y.
• format_fn – Function to format raw data to model input.
• overfit – If only one batch should be used thereby causing overfitting.

on_epoch_end() → None

Shuffle data after every epoch.

class deepblink.datasets.SpotsDataset(name: str, cell_size: int, smooth_factor: float = 1)

Bases: deepblink.datasets._datasets.Dataset

Class used to load all spots data.

Parameters:

• cell_size – Number of pixels (from original image) constituting one cell in the prediction matrix.
• smooth_factor – Value used to weigh true cells, weighs false cells with 1-smooth_factor.

image_size

Check if all images have the same square shape.

load_data() → None

Load dataset into memory.

normalize_dataset() → None

Normalize all the images to have zero mean and standard deviation 1.

prepare_data() → None

Convert raw labels into labels usable for training.

In the “spots” format, training labels are stored as lists of coordinates, this format cannot be used for training. Here, this format is converted into prediction matrices.