Skip to content

random_channel

Random Channel Layer API

This module provides classes to build random channel layers.

Classes:

Classes

RandomChannel 

RandomChannel(batchwise: bool = False, **kwargs)

Randomly picks a single channel from the input samples.

Parameters:

  • batchwise (bool, default: False ) –

    If True, grabs same channel from all samples in batch.

Source code in neuralspot_edge/layers/preprocessing/random_channel.py 
21
22
23
24
25
26
27
28
29
def __init__(self, batchwise: bool = False, **kwargs):
    """Randomly picks a single channel from the input samples.

    Args:
        batchwise (bool): If True, grabs same channel from all samples in batch.

    """
    super().__init__(**kwargs)
    self.batchwise = batchwise

Functions

get_random_transformations 
get_random_transformations(input_shape)

Generate random channel index.

Source code in neuralspot_edge/layers/preprocessing/random_channel.py 
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
def get_random_transformations(self, input_shape):
    """Generate random channel index."""

    batch = input_shape[0]
    if self.data_format == "channels_first":
        ch_size = input_shape[self.ch_axis]
    else:
        ch_size = input_shape[self.ch_axis]

    # Pick same channel for all samples in the batch
    if self.batchwise:
        ch_idx = keras.random.randint(shape=(), minval=0, maxval=ch_size, seed=self.random_generator, dtype="int32")
        ch_idx = keras.ops.broadcast_to(ch_idx, [batch])
    else:
        ch_idx = keras.random.randint(
            shape=(batch,), minval=0, maxval=ch_size, seed=self.random_generator, dtype="int32"
        )
    return {"channel": ch_idx}
augment_samples 
augment_samples(inputs)

Augment samples

Source code in neuralspot_edge/layers/preprocessing/random_channel.py 
50
51
52
53
54
55
56
57
58
59
60
def augment_samples(self, inputs):
    """Augment samples"""
    # If batchwise, easier to grab the channel here
    if self.batchwise:
        samples = inputs[self.SAMPLES]
        channel = inputs[self.TRANSFORMS]["channel"]
        channel = keras.ops.take(channel, 0, axis=0)
        return keras.ops.expand_dims(keras.ops.take(samples, channel, axis=self.ch_axis), axis=self.ch_axis)
    # Otherwise let the augment_sample method handle it
    else:
        super().augment_samples(inputs)
augment_sample 
augment_sample(inputs) -> keras.KerasTensor

Augment a sample during training.

Source code in neuralspot_edge/layers/preprocessing/random_channel.py 
62
63
64
65
66
def augment_sample(self, inputs) -> keras.KerasTensor:
    """Augment a sample during training."""
    sample = inputs[self.SAMPLES]
    channel = inputs[self.TRANSFORMS]["channel"]
    keras.ops.expand_dims(keras.ops.take(sample, channel, axis=self.ch_axis), axis=self.ch_axis)