extract

% grt-extract % %

NANE

grt-extract - extract features from a data sequence

SYNOPSIS

grt extract [-v|--verbose <1..4>] [-h|--help] [-q|--no-header] <feature-vector> [input-file]

grt extract list

DESCRIPTION

This program implements common feature extraction methods found in the literature. Supplying 'list' to the extract program will list all available extractors. Extract will work on the standard input format of the grtool suite. Each line encompasses a feature, made of a label and n sensor readings. An empty line designates the end of a fragment. Each feature extractor that you choose will be applied to a fragment. Since all extracctors are aggregating functions, only a single line will be returned, labelled with the first label in the fragment.

-h, --help : Print a help message. If an extractor is specified, the option for this extractor will be printed also.

-v, --verbose <0..4> : Tell the command to be more verbose about its execution.

-q, --no-header : Do not print the optional header line in the output

EXAMPLES

For starters let's list all available extraction modules:

grt extract list
usage: extract [options] ... <feature-extractor>
options:
  -v, --verbose        verbosity level: 0-4 (int [=0])
  -h, --help           print this message
  -q, --no-header      do not print the header
  -z, --z-normalize    z-normalize ( (x-mean(x))/std(x) ) all samples
  -o, --o-normalize    o-normalize, compute x_i - x_0, i.e. remove the first component from each sample

Available Extractors:

 mean (m): compute mean/average of each axis
 range (r): compute range (min/max) and their difference
 variance (v): compute variance of each axis
 median (e): compute median of each axis
 zcr (z): zero-crossing rate
 rms (s): root-mean squared over each and all axis
 time (t): shorthand for all time-domain features: mean,variance,range,median

We could then decide to extract mean and range from our input. This would result in the following command line:

echo "inverting 1 1
> inverting 0 0
> pipetting 2 3
> pipetting 2 2" | grt extract m range
# mean	range	
pipetting	1.25	1.5	2	0	2	3	0	3	

For each input dimension, the feature will be calculated and printed on the output. The mean extractor returns two dimensions, while the range extractor gives the maximum, minimum and the span of each input axis. This results in an eight-dimensional feature vector as output.

So, this aggregates multiple segments into one frame. Frames are separated by empty lines into segments. The following examples will aggregate the two segments into two frames!

echo "inverting 1 1
> inverting 0 0
>
> pipetting 2 3
> pipetting 2 2 " | grt e m
# mean	
inverting	0.5	0.5	
pipetting	2	2.5