Function for display of rocs (receiver operator characteristic curves).
Display roc curves. Consistent usage ensures uniform look for rocs. The
input D should have n rows, each of which is of the form:
D = [falsePosRate truePosRate]
D is generated, for example, by scanning a detection threshold over n
values from 0 (so first entry is [1 1]) to 1 (so last entry is [0 0]).
Alternatively D can be a cell vector of rocs, in which case an average
ROC will be shown with error bars. Plots missRate (which is just 1 minus
the truePosRate) on the y-axis versus the falsePosRate on the x-axis.
USAGE
[h,miss,stds] = plotRoc( D, prm )
INPUTS
D - [nx2] n data points along roc [falsePosRate truePosRate]
typically ranges from [1 1] to [0 0] (or may be reversed)
prm - [] param struct
.color - ['g'] color for curve
.lineSt - ['-'] linestyle (see LineSpec)
.lineWd - [4] curve width
.logx - [0] use logarithmic scale for x-axis
.logy - [0] use logarithmic scale for y-axis
.marker - [''] marker type (see LineSpec)
.mrkrSiz - [12] marker size
.nMarker - [5] number of markers (regularly spaced) to display
.lims - [0 1 0 1] axes limits
.smooth - [0] if T compute lower envelop of roc to smooth staircase
.fpTarget - [] return miss rates at given fp values (and draw lines)
.xLbl - ['false positive rate'] label for x-axis
.yLbl - ['miss rate'] label for y-axis
OUTPUTS
h - plot handle for use in legend only
miss - average miss rates at fpTarget reference values
stds - standard deviation of miss rates at fpTarget reference values
EXAMPLE
k=2; x=0:.0001:1; data1 = [1-x; (1-x.^k).^(1/k)]';
k=3; x=0:.0001:1; data2 = [1-x; (1-x.^k).^(1/k)]';
hs(1)=plotRoc(data1,struct('color','g','marker','s'));
hs(2)=plotRoc(data2,struct('color','b','lineSt','--'));
legend( hs, {'roc1','roc2'} ); xlabel('fp'); ylabel('fn');
See also
Piotr's Computer Vision Matlab Toolbox Version 3.02
Copyright 2014 Piotr Dollar. [pdollar-at-gmail.com]
Licensed under the Simplified BSD License [see external/bsd.txt]