qualia_plugin_snn.preprocessing package

Submodules

• qualia_plugin_snn.preprocessing.Group2TimeStepsBySample module
  • Group2TimeStepsBySample
    • Group2TimeStepsBySample.__init__()
    • Group2TimeStepsBySample.__call__()
• qualia_plugin_snn.preprocessing.IntegrateEventsByFixedDuration module
  • IntegrateEventsByFixedDuration
    • IntegrateEventsByFixedDuration.__init__()
    • IntegrateEventsByFixedDuration.__call__()
    • IntegrateEventsByFixedDuration.import_data()
• qualia_plugin_snn.preprocessing.IntegrateEventsByFixedFramesNumber module
  • IntegrateEventsByFixedFramesNumber
    • IntegrateEventsByFixedFramesNumber.__init__()
    • IntegrateEventsByFixedFramesNumber.__call__()
    • IntegrateEventsByFixedFramesNumber.import_data()
• qualia_plugin_snn.preprocessing.Split2TimeSteps module
  • Split2TimeSteps
    • Split2TimeSteps.__init__()
    • Split2TimeSteps.__call__()

Module contents

Qualia-Plugin-SNN preprocessing package contains preprocessing modules adapted for or dedicated to Spiking Neural Networks.

class qualia_plugin_snn.preprocessing.Group2TimeStepsBySample

Bases: Preprocessing[RawDataModel, RawDataModel]

Preprocessing module to group frame data from same sample into timesteps.

__call__(datamodel: RawDataModel) → RawDataModel

Group frames from the same samples of from a qualia_core.datamodel.RawDataModel.RawDataModel into timesteps.

Relies on sample indices (begin, end) from the info array to only group frames from the same sample.

Input data should be 2D or 1D (+ channel) with [N, H, W, C] order (channels_last). Output data has [N // T, T, H, W, C] dimensions Extra data from a sample that do not fit in a timestep group is truncated.

Parameters:

datamodel (RawDataModel) – The input dataset

Returns:

The dataset with additional timestep dimension

Return type:

RawDataModel

__init__(timesteps: int) → None

Construct qualia_plugin_snn.preprocessing.Group2TimeStepsBySample.Group2TimeStepsBySample.

Parameters:

timesteps (int) – Number of timesteps to group frames from a sample

Return type:

None

class qualia_plugin_snn.preprocessing.IntegrateEventsByFixedDuration

Bases: Preprocessing[EventDataModel, RawDataModel]

Preprocessing module to construct fixed-duration frames from event data.

__call__(datamodel: EventDataModel) → RawDataModel

Construct frames from events of the same sample of a qualia_plugin_snn.datamodel.EventDataModel.EventDataModel.

Relies on sample indices (begin, end) from the info array to only collect events from the same sample.

Input data should be 2D event data with (t, x, y, p) columns or 1D event data with (t, x, p) columns. Output data has [N, H, W, C] or [N, W, C] dimensions

Uses SpikingJelly’s implementation of spikingjelly.datasets.integrate_events_segment_to_frame()

Parameters:

datamodel (EventDataModel) – The input event-based dataset

Returns:

The new frame dataset

Return type:

RawDataModel

__init__(duration: int) → None

Construct qualia_plugin_snn.preprocessing.IntegrateEventsByFixedDuration.IntegrateEventsByFixedDuration.

Parameters:

duration (int) – Duration of frame

Return type:

None

import_data(dataset: Dataset[Any]) → Dataset[Any]

Parameters:

dataset (Dataset[Any])

Return type:

Dataset[Any]

class qualia_plugin_snn.preprocessing.IntegrateEventsByFixedFramesNumber

Bases: IntegrateEventsByFixedDuration

Preprocessing module to construct fixed number of frames from a sample of event data.

__call__(datamodel: EventDataModel) → RawDataModel

Construct frames from events of the same sample of a qualia_plugin_snn.datamodel.EventDataModel.EventDataModel.

Relies on sample indices (begin, end) from the info array to only collect events from the same sample.

Input data should be 2D event data with (t, x, y, p) columns or 1D event data with (t, x, p) columns. Output data has [N, H, W, C] or [N, W, C] dimensions

Uses a derivative of SpikingJelly’s implementation of spikingjelly.datasets.integrate_events_by_fixed_frames_number()

Parameters:

datamodel (EventDataModel) – The input event-based dataset

Returns:

The new frame dataset

Return type:

RawDataModel

__init__(split_by: str, frames_num: int) → None

Construct qualia_plugin_snn.preprocessing.IntegrateEventsByFixedFramesNumber.IntegrateEventsByFixedFramesNumber.

Parameters:

• split_by (str) – 'time' or 'number', see spikingjelly.datasets.cal_fixed_frames_number_segment_index()

• frames_num (int) – Number of frames to produce per sample

Return type:

None

import_data(dataset: Dataset[Any]) → Dataset[Any]

Parameters:

dataset (Dataset[Any])

Return type:

Dataset[Any]

class qualia_plugin_snn.preprocessing.Split2TimeSteps

Bases: Preprocessing[RawDataModel, RawDataModel]

Preprocessing module to split 1D input dataset into multiple timesteps.

__call__(datamodel: RawDataModel) → RawDataModel

Split the given qualia_core.datamodel.RawDataModel.RawDataModel into multiple timesteps.

Input data should be 1D (+ channel) with [N, S, C] order (channels_last). Output data has [N, T, S // T, C] dimensions Extra data that do not fit in a chunk is truncated.

Parameters:

datamodel (RawDataModel) – The input dataset

Returns:

The dataset with additional timestep dimension

Return type:

RawDataModel

__init__(chunks: int) → None

Construct qualia_plugin_snn.preprocessing.Split2TimeSteps.Split2TimeSteps.

Parameters:

chunks (int) – Number of chunks to split the data into

Return type:

None