adaptive_threshold_classifier¶

Module: `missions.nodes.classification.adaptive_threshold_classifier`¶

Setting adaptive threshold

Inheritance diagram for pySPACE.missions.nodes.classification.adaptive_threshold_classifier:

Inheritance diagram of pySPACE.missions.nodes.classification.adaptive_threshold_classifier

Class Summary¶

`AdaptiveThresholdPreprocessingNode`([...])	Setting adaptive threshold as described by Semmaoui, H., etal.
`AdaptiveThresholdClassifierNode`([...])	Adaptive threshold onset detection classifier

Classes¶

`AdaptiveThresholdPreprocessingNode`¶

class pySPACE.missions.nodes.classification.adaptive_threshold_classifier.AdaptiveThresholdPreprocessingNode(width_adaptive_threshold=2000, p_adaptive_threshold=8, time_below_threshold=1000, **kwargs)[source]¶

Bases: pySPACE.missions.nodes.base_node.BaseNode

Setting adaptive threshold as described by Semmaoui, H., etal. (2012)

This node can be used to threshold a continuous signal with a adaptive threshold. The advantage over a simple fixed threshold method is the adaption to the signal. For example if a sensor value drifts over time either in positive or negative direction, a fixed threshold method can have big problems with this one. For a negative drift the “zero” value may get so low that the fixed threshold is never reached again, the other way round a positive drift can lead to a continuous overcoming of the fixed threshold. The adaptive threshold is based on the following publication:

Semmaoui, H., etal. (2012). Setting adaptive spike detection threshold for smoothed TEO based on robust statistics theory. IEEE Transactions on Biomedical Engineering, 59(2):474 - 482. (http://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=06070974)

The formula is given as:

$T(t) = mean(t)_N + p*std(t)_N,$

where T(t) is the threshold at a given timepoint t, mean(t)_N is the mean at timepoint t calculated over the last N samples p is the sensitivity factor and std(t)_N is the standard deviation at timepoint t calculated over the last N samples.

The processing is split into two parts, this node implements the first part which does the actual thresholding and safes a timeseries containing zeros, accept at those timepoints where the signal exceeded the threshold. NOTICE only the very first timepoint where the signal overcame the threshold is marked with the value 1 all other values remain a zero. In a second step see “AdaptiveThresholdClassifierNode” below the results are transfered into prediction vectors. This is done since the threshold methods needs the whole data in order to continuously calculate the mean and std. dev., otherwise the first N samples of each window could not be used for analysis. IMPORTANT the preprocessing has to be done without any windowing accept the NULL marker, with a fixed nullmarkerstride.

Parameters

width_adaptiveThreshold:

Specifies the width of the window used for calculating the mean and the standard deviation for the threshold in ms

(optional, default:2000)

p_adaptiveThreshold:

Specifies the p for the adaptive threshold

(optional, default:8)

time_below_threshold:

Specifies how long the signal has to be below the signal before a new thresholding is allowed in ms. This is helpful if only the beginning of some event should be detected in the signal.

(optional, default:1000)

Exemplary Call

-
    node : AdaptiveThreshold_Preprocessing
    parameters :
        width_adaptive_threshold : 2000
        p_adaptive_threshold : 8
        time_below_threshold : 1000

POSSIBLE NODE NAMES:
Author:	Marc Tabie (mtabie@informatik.uni-bremen.de)
Created:	2013/01/17
Last change:	2013/01/23 by Marc Tabie
	AdaptiveThreshold_Preprocessing AdaptiveThresholdPreprocessingNode AdaptiveThresholdPreprocessing
POSSIBLE INPUT TYPES:
	TimeSeries

Class Components Summary

`_execute`(x)	Executes the preprocessing on the given data vector x
`adaptive_threshold`(data_point, channel_counter)	Adaptive threshold for single values
`get_output_type`(input_type[, as_string])
`input_types`
`is_supervised`()	Returns whether this node requires supervised training
`is_trainable`()	Returns whether this node is trainable.

input_types = ['TimeSeries']¶

__init__(width_adaptive_threshold=2000, p_adaptive_threshold=8, time_below_threshold=1000, **kwargs)[source]¶

is_trainable()[source]¶: Returns whether this node is trainable.

is_supervised()[source]¶: Returns whether this node requires supervised training

_execute(x)[source]¶: Executes the preprocessing on the given data vector x

get_output_type(input_type, as_string=True)[source]¶

adaptive_threshold(data_point, channel_counter)[source]¶

Adaptive threshold for single values

data_point = new datapoint channel_counter = index for the retained channels in the ringbuffer