source: python/test/bench/onset/benchonset.py @ ec10780

#! /usr/bin/python

from aubio.bench.node import *
from os.path import dirname,basename

def mmean(l):
        return sum(l)/max(float(len(l)),1)

def stdev(l):
        smean = 0
        if not len(l): return smean
        lmean = mmean(l)
        for i in l:
                smean += (i-lmean)**2
        smean *= 1. / len(l)
        return smean**.5

class benchonset(bench):

        """ list of values to store per file """
        valuenames = ['orig','missed','Tm','expc','bad','Td']
        """ list of lists to store per file """
        valuelists = ['l','labs']
        """ list of values to print per dir """
        printnames = [ 'mode', 'thres', 'dist', 'prec', 'recl',
                'GD', 'FP', 
                'Torig', 'Ttrue', 'Tfp',  'Tfn',  'TTm',   'TTd',
                'aTtrue', 'aTfp', 'aTfn', 'aTm',  'aTd',  
                'mean', 'smean',  'amean', 'samean']

        """ per dir """
        formats = {'mode': "%12s" , 'thres': "%5.4s", 
                'dist':  "%5.4s", 'prec': "%5.4s", 'recl':  "%5.4s",
                'Torig': "%5.4s", 'Ttrue': "%5.4s", 'Tfp':   "%5.4s", 'Tfn':   "%5.4s", 
                'TTm':    "%5.4s", 'TTd':    "%5.4s",
                'aTtrue':"%5.4s", 'aTfp':  "%5.4s", 'aTfn':  "%5.4s", 
                'aTm':   "%5.4s", 'aTd':   "%5.4s",
                'mean':  "%5.40s", 'smean': "%5.40s", 
                'amean':  "%5.40s", 'samean': "%5.40s", 
                "GD":     "%5.4s", "FP":     "%5.4s",
                "GDm":     "%5.4s", "FPd":     "%5.4s"}

        def dir_eval(self):
                """ evaluate statistical data over the directory """
                v = self.v

                v['mode']      = self.params.onsetmode
                v['thres']     = self.params.threshold 
                v['bufsize']   = self.params.bufsize
                v['hopsize']   = self.params.hopsize
                v['silence']   = self.params.silence
                v['mintol']   = self.params.mintol

                v['Torig']     = sum(v['orig'])
                v['TTm']       = sum(v['Tm'])
                v['TTd']       = sum(v['Td'])
                v['Texpc']     = sum(v['expc'])
                v['Tbad']      = sum(v['bad'])
                v['Tmissed']   = sum(v['missed'])
                v['aTm']       = mmean(v['Tm'])
                v['aTd']       = mmean(v['Td'])

                v['mean']      = mmean(v['l'])
                v['smean']     = stdev(v['l'])

                v['amean']     = mmean(v['labs'])
                v['samean']    = stdev(v['labs'])
                
                # old type calculations
                # good detection rate 
                v['GD']  = 100.*(v['Torig']-v['Tmissed']-v['TTm'])/v['Torig']
                # false positive rate
                v['FP']  = 100.*(v['Tbad']+v['TTd'])/v['Torig']
                # good detection counting merged detections as good
                v['GDm'] = 100.*(v['Torig']-v['Tmissed'])/v['Torig'] 
                # false positives counting doubled as good
                v['FPd'] = 100.*v['Tbad']/v['Torig']                
                
                # mirex type annotations
                totaltrue = v['Texpc']-v['Tbad']-v['TTd']
                totalfp = v['Tbad']+v['TTd']
                totalfn = v['Tmissed']+v['TTm']
                self.v['Ttrue']     = totaltrue
                self.v['Tfp']       = totalfp
                self.v['Tfn']       = totalfn
                # average over the number of annotation files
                N = float(len(self.reslist))
                self.v['aTtrue']    = totaltrue/N
                self.v['aTfp']      = totalfp/N
                self.v['aTfn']      = totalfn/N

                # F-measure
                self.P = 100.*float(totaltrue)/max(totaltrue + totalfp,1)
                self.R = 100.*float(totaltrue)/max(totaltrue + totalfn,1)
                #if self.R < 0: self.R = 0
                self.F = 2.* self.P*self.R / max(float(self.P+self.R),1)
                self.v['dist']      = self.F
                self.v['prec']      = self.P
                self.v['recl']      = self.R


        """
        Plot functions 
        """

        def plotroc(self,d,plottitle=""):
                import Gnuplot, Gnuplot.funcutils
                gd = []
                fp = []
                for i in self.vlist:
                        gd.append(i['GD']) 
                        fp.append(i['FP']) 
                d.append(Gnuplot.Data(fp, gd, with='linespoints', 
                        title="%s %s" % (plottitle,i['mode']) ))

        def plotplotroc(self,d,outplot=0,extension='ps'):
                import Gnuplot, Gnuplot.funcutils
                from sys import exit
                g = Gnuplot.Gnuplot(debug=0, persist=1)
                if outplot:
                        if   extension == 'ps':  ext, extension = '.ps' , 'postscript'
                        elif extension == 'png': ext, extension = '.png', 'png'
                        elif extension == 'svg': ext, extension = '.svg', 'svg'
                        else: exit("ERR: unknown plot extension")
                        g('set terminal %s' % extension)
                        g('set output \'roc-%s%s\'' % (outplot,ext))
                xmax = 30 #max(fp)
                ymin = 50 
                g('set xrange [0:%f]' % xmax)
                g('set yrange [%f:100]' % ymin)
                # grid set
                g('set grid')
                g('set xtics 0,5,%f' % xmax)
                g('set ytics %f,5,100' % ymin)
                g('set key 27,65')
                #g('set format \"%g\"')
                g.title(basename(self.datadir))
                g.xlabel('false positives (%)')
                g.ylabel('correct detections (%)')
                g.plot(*d)

        def plotpr(self,d,plottitle=""):
                import Gnuplot, Gnuplot.funcutils
                x = []
                y = []
                for i in self.vlist:
                        x.append(i['prec']) 
                        y.append(i['recl']) 
                d.append(Gnuplot.Data(x, y, with='linespoints', 
                        title="%s %s" % (plottitle,i['mode']) ))

        def plotplotpr(self,d,outplot=0,extension='ps'):
                import Gnuplot, Gnuplot.funcutils
                from sys import exit
                g = Gnuplot.Gnuplot(debug=0, persist=1)
                if outplot:
                        if   extension == 'ps':  ext, extension = '.ps' , 'postscript'
                        elif extension == 'png': ext, extension = '.png', 'png'
                        elif extension == 'svg': ext, extension = '.svg', 'svg'
                        else: exit("ERR: unknown plot extension")
                        g('set terminal %s' % extension)
                        g('set output \'pr-%s%s\'' % (outplot,ext))
                g.title(basename(self.datadir))
                g.xlabel('Recall (%)')
                g.ylabel('Precision (%)')
                g.plot(*d)

        def plotfmeas(self,d,plottitle=""):
                import Gnuplot, Gnuplot.funcutils
                x,y = [],[]
                for i in self.vlist:
                        x.append(i['thres']) 
                        y.append(i['dist']) 
                d.append(Gnuplot.Data(x, y, with='linespoints', 
                        title="%s %s" % (plottitle,i['mode']) ))

        def plotplotfmeas(self,d,outplot="",extension='ps', title="F-measure"):
                import Gnuplot, Gnuplot.funcutils
                from sys import exit
                g = Gnuplot.Gnuplot(debug=0, persist=1)
                if outplot:
                        if   extension == 'ps':  terminal = 'postscript'
                        elif extension == 'png': terminal = 'png'
                        elif extension == 'svg': terminal = 'svg'
                        else: exit("ERR: unknown plot extension")
                        g('set terminal %s' % terminal)
                        g('set output \'fmeas-%s.%s\'' % (outplot,extension))
                g.xlabel('threshold \\delta')
                g.ylabel('F-measure (%)')
                g('set xrange [0:1.2]')
                g('set yrange [0:100]')
                g.title(basename(self.datadir))
                # grid set
                #g('set grid')
                #g('set xtics 0,5,%f' % xmax)
                #g('set ytics %f,5,100' % ymin)
                #g('set key 27,65')
                #g('set format \"%g\"')
                g.plot(*d)

        def plotfmeasvar(self,d,var,plottitle=""):
                import Gnuplot, Gnuplot.funcutils
                x,y = [],[]
                for i in self.vlist:
                        x.append(i[var]) 
                        y.append(i['dist']) 
                d.append(Gnuplot.Data(x, y, with='linespoints', 
                        title="%s %s" % (plottitle,i['mode']) ))
        
        def plotplotfmeasvar(self,d,var,outplot="",extension='ps', title="F-measure"):
                import Gnuplot, Gnuplot.funcutils
                from sys import exit
                g = Gnuplot.Gnuplot(debug=0, persist=1)
                if outplot:
                        if   extension == 'ps':  terminal = 'postscript'
                        elif extension == 'png': terminal = 'png'
                        elif extension == 'svg': terminal = 'svg'
                        else: exit("ERR: unknown plot extension")
                        g('set terminal %s' % terminal)
                        g('set output \'fmeas-%s.%s\'' % (outplot,extension))
                g.xlabel(var)
                g.ylabel('F-measure (%)')
                #g('set xrange [0:1.2]')
                g('set yrange [0:100]')
                g.title(basename(self.datadir))
                g.plot(*d)

        def plotdiffs(self,d,plottitle=""):
                import Gnuplot, Gnuplot.funcutils
                v = self.v
                l = v['l']
                mean   = v['mean']
                smean  = v['smean']
                amean  = v['amean']
                samean = v['samean']
                val = []
                per = [0] * 100
                for i in range(0,100):
                        val.append(i*.001-.05)
                        for j in l: 
                                if abs(j-val[i]) <= 0.001:
                                        per[i] += 1
                total = v['Torig']
                for i in range(len(per)): per[i] /= total/100.

                d.append(Gnuplot.Data(val, per, with='fsteps', 
                        title="%s %s" % (plottitle,v['mode']) ))
                #d.append('mean=%f,sigma=%f,eps(x) title \"\"'% (mean,smean))
                #d.append('mean=%f,sigma=%f,eps(x) title \"\"'% (amean,samean))


        def plotplotdiffs(self,d,outplot=0,extension='ps'):
                import Gnuplot, Gnuplot.funcutils
                from sys import exit
                g = Gnuplot.Gnuplot(debug=0, persist=1)
                if outplot:
                        if   extension == 'ps':  ext, extension = '.ps' , 'postscript'
                        elif extension == 'png': ext, extension = '.png', 'png'
                        elif extension == 'svg': ext, extension = '.svg', 'svg'
                        else: exit("ERR: unknown plot extension")
                        g('set terminal %s' % extension)
                        g('set output \'diffhist-%s%s\'' % (outplot,ext))
                g('eps(x) = 1./(sigma*(2.*3.14159)**.5) * exp ( - ( x - mean ) ** 2. / ( 2. * sigma ** 2. ))')
                g.title(basename(self.datadir))
                g.xlabel('delay to hand-labelled onset (s)')
                g.ylabel('% number of correct detections / ms ')
                g('set xrange [-0.05:0.05]')
                g('set yrange [0:20]')
                g.plot(*d)


        def plothistcat(self,d,plottitle=""):
                import Gnuplot, Gnuplot.funcutils
                total = v['Torig']
                for i in range(len(per)): per[i] /= total/100.

                d.append(Gnuplot.Data(val, per, with='fsteps', 
                        title="%s %s" % (plottitle,v['mode']) ))
                #d.append('mean=%f,sigma=%f,eps(x) title \"\"'% (mean,smean))
                #d.append('mean=%f,sigma=%f,eps(x) title \"\"'% (amean,samean))


        def plotplothistcat(self,d,outplot=0,extension='ps'):
                import Gnuplot, Gnuplot.funcutils
                from sys import exit
                g = Gnuplot.Gnuplot(debug=0, persist=1)
                if outplot:
                        if   extension == 'ps':  ext, extension = '.ps' , 'postscript'
                        elif extension == 'png': ext, extension = '.png', 'png'
                        elif extension == 'svg': ext, extension = '.svg', 'svg'
                        else: exit("ERR: unknown plot extension")
                        g('set terminal %s' % extension)
                        g('set output \'diffhist-%s%s\'' % (outplot,ext))
                g('eps(x) = 1./(sigma*(2.*3.14159)**.5) * exp ( - ( x - mean ) ** 2. / ( 2. * sigma ** 2. ))')
                g.title(basename(self.datadir))
                g.xlabel('delay to hand-labelled onset (s)')
                g.ylabel('% number of correct detections / ms ')
                g('set xrange [-0.05:0.05]')
                g('set yrange [0:20]')
                g.plot(*d)