sksfa.utils.ReceptiveSlicer

sksfa.utils.ReceptiveSlicer(input_shape, field_size=(3, 3), strides=(1, 1), padding='valid', copy=True)

Slicing part of field slicing.

This transformer takes input of shape (n_samples, width, height, channels) and slices inputs in a receptive field manner.

Parameters

field_sizetuple

Shape of the receptive field as a tuple of integers.

stridestuple

Strides in each axis as tuple of integers.

paddingstr

Either “valid” or “same”. Only “valid” is implemented as of now.

copybool, default=True

If False, data passed to fit are overwritten and running fit(X).transform(X) will not yield the expected results, use fit_transform(X) instead.

Examples

>>> from sksfa.utils import ReceptiveSlicer
>>> import numpy as np
>>>
>>> ones = np.ones((2, 2))
>>> # This could be an image or rebuilt output from a lower layer
>>> data = np.block([[0 * ones, 1 * ones], [2 * ones, 3 * ones]])[None, ..., None]
>>>
>>> print(data.squeeze())
[[0. 0. 1. 1.]
 [0. 0. 1. 1.]
 [2. 2. 3. 3.]
 [2. 2. 3. 3.]]
>>> print(f"Input shape: {data.shape}")
Input shape: (1, 4, 4, 1)
>>> slicer = ReceptiveSlicer(input_shape=data.shape, field_size=ones.shape, strides=(1, 1))
>>> slicer = slicer.fit(data)
>>> sliced_output = slicer.transform(data)
>>> print(f"Output shape: {sliced_output.shape}")
Output shape: (9, 4)
>>> for idx, field_sample in enumerate(sliced_output): print(f"Output sample {idx}: {field_sample.squeeze()}")
Output sample 0: [0. 0. 0. 0.]
Output sample 1: [0. 1. 0. 1.]
Output sample 2: [1. 1. 1. 1.]
Output sample 3: [0. 0. 2. 2.]
Output sample 4: [0. 1. 2. 3.]
Output sample 5: [1. 1. 3. 3.]
Output sample 6: [2. 2. 2. 2.]
Output sample 7: [2. 3. 2. 3.]
Output sample 8: [3. 3. 3. 3.]