def plotSvm(event=0,suf=''):
''' plots the grid search results '''
print 'plotSvm started'
plt.figure()
infos=[]
for vp in [1,2,3,4]:
path,inpath,figpath=initPath(vp,event)
fns=os.listdir(inpath+'svm%s/'%suf)
fns=filter(lambda s: s.endswith('.log'),fns)
dat=[]
for fn in fns:
f=open(inpath+'svm%s/'%suf+fn,'r')
txt=f.read()
f.close()
txt='\n'+txt
txt=txt.rsplit('%')
for tx in txt[:-1]:
lines=tx.rsplit('\n')
#print len(lines),len(lines[0]),lines[0]
b= float(lines[1].rsplit('=')[1])
C= float(lines[2].rsplit('=')[1])
f= float(lines[-1].rsplit('=')[1])/100.
dat.append([b,C,f])
dat=np.array(dat)
betas=np.unique(dat[:,0]).tolist()
Cs= np.unique(dat[:,1]).tolist()
fun=np.zeros((len(betas),len(Cs)))#*np.nan
for d in dat.tolist():
fun[betas.index(d[0]),Cs.index(d[1])]=d[2]
inc=(betas[1]-betas[0])
betas.append(betas[-1]+inc)
Cs.append(Cs[-1]+inc)
betas=np.array(betas)-inc/2.;Cs=np.array(Cs)-inc/2.
am= (fun==np.max(fun)).nonzero()
iam=np.argmin(am[1])
opt=[betas[am[0][iam]]+inc/2.,Cs[am[1][iam]]+inc/2.]
# sanity check
oi=np.logical_and(opt[0]==dat[:,0],opt[1]==dat[:,1])
assert np.max(fun)==dat[oi.nonzero()[0][0],2]
np.save(inpath+'svm%s/opt'%suf,opt)
nf=getWeights(vp,event,suf)
chnc=max(nf[0],nf[1])/float(nf[0]+nf[1])
infos.append([vp,np.max(fun)*100,chnc*100]+nf.tolist()
+[nf[2]/float(nf[0]+nf[1])*100,opt[1],opt[0]])
plt.subplot(2,2,vp)
plt.pcolor(betas,Cs,fun.T,cmap='hot')
plt.xlabel('beta');plt.ylabel('C')
plt.xlim([betas[0],betas[-1]]);plt.ylim([Cs[0],Cs[-1]])
plt.colorbar()
plt.plot(opt[0],opt[1],'rx',mew=2)
plt.title('b=%.1f, C=%.1f,fm=%.2f,ch=%.2f'%(opt[0],opt[1],np.max(fun),chnc))
plt.savefig(figpath+'svm%sfitEv%d.png'%(suf,event))
from matustools.matusplotlib import ndarray2latextable
ndarray2latextable(np.array(infos),decim=[0,2,2,0,0,0,2,1,1])
return infos
def tabSampleSize():
res=[]
for vp in range(1,5):
res.append([])
for ev in [97,0,1]:
path=inpath+'vp%03d/E%d/'%(vp,ev)
sc=np.load(path+'X/score.npy')
res[-1].append(sc.shape[0])
ti=np.load(inpath+'vp%03d/'%vp+'ti.npy')
res[-1].append(ti.shape[0])
res=np.array(res)
ndarray2latextable(res,decim=0)
def railIdeal():
T=68; P=64
t=np.linspace(-0.8,0,T);p=np.linspace(-5,5,P)
fn=inpath+'vp%03d/E%d/X/coeff.npy'%(999,1)
pc=_getPC(np.load(fn),0)
if pc.mean()>=0.4: pc=1-pc
D= pc.T[:,31:33,:].mean(1)
D/=D.sum();
x0=np.array((3,-12,0.1,7,-12,0.1))
xopt=fmin(func=_fun,x0=x0,args=(D,t,p,False)).tolist()
xopt.append(abs(xopt[0]-xopt[3]))
ndarray2latextable(np.array(xopt,ndmin=2),decim=2)
def si2tex():
from matustools.matusplotlib import ndarray2latextable
res=[]
for vp in range(1,5):
vp,ev,path=initVP(vp,0)
si=np.load(path+'si.npy')
out=[]
for ev in range(6):
out.append(np.sum(si[:,14]==ev))
out.append(np.sum(si[:,14]>ev))
out.append(np.sum(si[:,13]==1))
out.extend([out[1]/float(out[0])*100, 100*out[-1]/float(out[1])])
res.append(out)
ndarray2latextable(np.array(res),8*[0]+[1,1])
def plotPC1rail():
T=68;P=64
t=np.linspace(-0.8,0.8,T);p=np.linspace(-5,5,P)
tm=np.repeat(t[np.newaxis,:],P,axis=0)
pm=np.repeat(p[:,np.newaxis],T,axis=1)
fig=figure(size=3,aspect=0.7)
fig.tight_layout()
bnds=[(1,None),(None,0),(None,None)];est=[]
for ev in [0,1]:
for vp in range(1,5):
fn=inpath+'vp%03d/E%d/'%(vp,ev)+'X/coeff.npy'
pc=_getPC(np.load(fn),0)
if pc.mean()>=0.4: pc=1-pc
inc=0
D= pc.T[:,(31+inc):(33+inc),:].mean(1)
D/=D.sum();
subplot(2,4,ev*4+vp)
plt.pcolor(p,t,D.T,cmap='gray')
# below we set the initial guess
x0=np.array((5,-12,0.2))
xopt=fmin(func=_fun,x0=x0,args=(D,t,p,False))
est.append(xopt.tolist())
plt.plot(xopt[0]-xopt[1]*t,t,'r',lw=1,alpha=0.4)
plt.grid(True,'both');
plt.xlim([p[0],p[-1]]);plt.ylim([t[0],t[-1]]);
ax=plt.gca();ax.set_axisbelow(False)
ax.set_xticks([-4,-2,0,2,4])
ax.set_yticks(np.linspace(-0.8,0.8,5))
if not ev: ax.set_xticklabels([])
if vp>1: ax.set_yticklabels([])
else:
ax.set_yticklabels(np.linspace(-0.8,0.8,5))
plt.ylabel(['ES','CS1'][ev]+'\nTime to saccade in sec.')
#if i==1: plt.text(2,-1,FIG[3][2],size=8)
#if i==1: plt.xlabel(FIG[3][2])
#else: plt.ylabel('subject %d'%(i+1))
if not ev: plt.title(FIG[3][3]+str(vp))
print FIG[2][1][0]
plt.subplot(2,4,6);plt.text(-9,-1.15,FIG[2][1][0])
plt.subplots_adjust(wspace=-1)
plt.savefig(figpath+'Pixel'+os.path.sep+'pcSacRail',
dpi=DPI,bbox_inches='tight')
est=np.array(est)
print est.ndim, est
if est.ndim>2: est=np.squeeze(est)
est[:,1]*=(t[-1]-t[0])/0.8
print ndarray2latextable(est,decim=2)
print np.corrcoef(est[:,1],est[:,2])[0,1]
def tabLatent(ev,pcs=5):
dat=[]
if ev==1: vps=range(1,5)+[999]
else: vps=range(1,5)
for vp in vps:
path=inpath+'vp%03d/E%d/X/'%(vp,ev)
dat.append(np.load(path+'latent.npy')[:pcs]*100)
return ndarray2latextable(np.array(dat),decim=1)
def tabLatent(ev,pcs=5):
'''pcs - number of principal components that will be displayed'''
dat=[]
if ev==1: vps=range(1,5)+[999]
else: vps=range(1,5)
for vp in vps:
path=inpath+'vp%03d/E%d/X/'%(vp,ev)
dat.append(np.load(path+'latent.npy')[:pcs]*100)
return ndarray2latextable(np.array(dat),decim=1)
def codingComparison():
res=[];
for vp in [2,1]:
initVP(vpl=vp,evl=0)
si1=np.load(path+'sicoder1.npy').tolist()
si2=np.load(path+'sicoder2.npy').tolist()
sel=[]
for i in range(len(si1)):
for j in range(i+1,len(si1)):
if si1[i][0]==si1[j][0] and si1[i][-1]==si1[j][-1] and si1[i][-2]==si1[j][-2]:
sel.append(j)
print sel
j=-1
for a in si1:
j+=1
if j in sel: continue
for b in si2:
if a[0]==b[0] and a[-1]==b[-1] and a[-2]==b[-2]: res.append([a[14],b[14]])
print vp,len(res),len(si1),len(si2)
def analyze(dat):
tp=np.logical_and(dat[:,0],dat[:,1]).sum()
fn=np.logical_and(~dat[:,0],dat[:,1]).sum()
tn=np.logical_and(~dat[:,0],~dat[:,1]).sum()
fp=np.logical_and(dat[:,0],~dat[:,1]).sum()
acc=(tn+tp)/float(dat.shape[0])
sens=tp/float(tp+fn)
spec=tn/float(fp+tn)
corr=(tp*tn-fp*fn)/np.sqrt((tp+fp)*(tp+fn)*(tn+fp)*(tn+fn))
return [acc,sens,spec,corr]
#np.save(path+'ress',res)
out=[];res=np.array(res)
for dat in [res>0,res==0,res==1]: out.append(analyze(dat))
res=res[np.logical_or(res[:,1]==0,res[:,1]==1),:]
out.append(analyze(res==1))
np.save(path+'codingComparison',out)
from matustools.matusplotlib import ndarray2latextable
print ndarray2latextable(np.array(out))
def plotBTpc(vpn=range(1,5),pcaEv=97):
''' vpn - list with subject ids
pcaEv - id of the catch-up saccade, 0=exploration saccade,
97 - 200 ms before button press'''
#dat=plotBTavg(MOVIE=False)
T=68#dat[0][0].shape[-1]
P=64#dat[0][0].shape[0]
t=np.linspace(-0.8,0,T);p=np.linspace(-5,5,P)
tm=np.repeat(t[np.newaxis,:],P,axis=0)
pm=np.repeat(p[:,np.newaxis],T,axis=1)
rows=len(vpn)
#cols=len(dat[0])
fig=figure(size=2,aspect=0.75)
fig.tight_layout()
#m=[-251,-201,-151,-101,-51,-1]
bnds=[(1,None),(None,0),(None,None)]
est=[]
for i in range(rows+2):
#for k in [1]:
#est.append([])
#j=4# 200 ms
fn=inpath+'vp%03d/E%d/'%(vpn[max(0,i-2)],pcaEv)+'X/coeff.npy'
if i==1: pc=(_getPC(np.load(fn),0)+_getPC(np.load(fn),1))/2.
elif i==0: pc=(_getPC(np.load(fn),0)-_getPC(np.load(fn),1)+1)/2.
else: pc=_getPC(np.load(fn),0)
if pc.mean()>=0.4: pc=1-pc
inc=[-2,2,0,0,0,0][i]
D= pc.T[:,(31+inc):(33+inc),:].mean(1)
D/=D.sum();
subplot(2,3,i+1)
plt.pcolor(p,t,D.T,cmap='gray')
# below we set the initial guess
if i==0: x0=np.array((1,-12,0.3,-2,-12,0.1))
elif i==1:x0=np.array((3,-12,0.1,0,-12,0.1))
elif i==2: x0=np.array((0,-12,0.1))
else: x0=np.array((3,-12,0.1,-2,-12,0.1))
xopt=fmin(func=_fun,x0=x0,args=(D,t,p,False))
est.append(xopt.tolist())
plt.plot(xopt[0]-xopt[1]*t,t,'r',lw=1,alpha=0.4)
if x0.size==6: plt.plot(xopt[3]-xopt[4]*t,t,'r',lw=1,alpha=0.4)
elif x0.size==3:est[-1].extend([np.nan]*3)
else: raise ValueError
est[-1].append(abs(est[-1][0]-est[-1][3]))
plt.grid(True,'both');
plt.xlim([p[0],p[-1]]);plt.ylim([t[0],t[-1]]);
ax=plt.gca();ax.set_axisbelow(False)
ax.set_xticks([-4,-2,0,2,4])
ax.set_yticks(np.linspace(-0.8,0,5))
if i%3!=0: ax.set_yticklabels([])
else:
ax.set_yticklabels(np.linspace(-1,-0.2,5))
if i<3: ax.set_xticklabels([])
if i==0: plt.text(-9,-0.3,FIG[3][1],size=8,rotation='vertical')
if i==4: plt.xlabel(FIG[3][2])
#else: plt.ylabel('subject %d'%(i+1))
#if i==0: plt.title(str(m[j]*2+2))
plt.text(2.1,-0.75,FIG[3][3]+str(vpn[max(i-2,0)])+['a','b',''][min(i,2)],color='w')
plt.subplots_adjust(wspace=-1)
plt.savefig(figpath+FIG[3][0]+'%d'%pcaEv,
dpi=DPI,bbox_inches='tight')
est=np.array(est)
#for k in [2,5]: est[:,k]=est[:,k]/np.sin(np.arctan(1/-est[:,k-1]))
print est.shape
if est.ndim>2: est=np.squeeze(est)
print ndarray2latextable(est,decim=2)
def plotBTavg(MAX=16,MOVIE=True):
from matustools.matusplotlib import plotGifGrid
dat=[];T=68;P=64;est=[]
t=np.linspace(-0.8,0,T);p=np.linspace(-5,5,P)
figure(size=3,aspect=1)
for vp in range(1,5):
dat.append([])
for event in range(-6,0):
fn=inpath+'vp%03d/E%d/'%(vp,100+event)+'PF/PF000.npy'
d=np.squeeze(np.load(fn))
#print np.max(d.mean(axis=0)),np.min(d.mean(axis=0))
dat[-1].append(d.mean(axis=0)/float(MAX))
inc=0#[-2,2,0,0,0,0][i+1]
D= (d.mean(axis=0)/float(MAX))
D=np.rollaxis(D,1)
D=D[:,(31+inc):(33+inc),:].mean(1);D/=D.sum();
j=vp-1; i=event+6
subplot(6,4,i*4+j+1)
plt.pcolor(p,t,D.T,cmap='gray')
# below we set the initial guess
if i==3:
if vp==1: x0=np.array((0.5,-12,0.3))
else: x0=np.array((3,-12,0.2,-0.5,-12,0.2))
xopt=fmin(func=_fun,x0=x0,args=(D,t,p,False))
est.append(xopt.tolist())
plt.plot(xopt[0]-xopt[1]*t,t,'r',lw=1,alpha=0.4)
if vp!=1: plt.plot(xopt[3]-xopt[4]*t,t,'r',lw=1,alpha=0.4)
else:est[-1].extend([np.nan]*3)
plt.grid(True,'both');
plt.xlim([p[0],p[-1]]);plt.ylim([t[0],t[-1]]);
ax=plt.gca();ax.set_axisbelow(False)
ax.set_xticks([-4,-2,0,2,4])
ytck=np.linspace(-0.8,0,5)[1:]
ax.set_yticks(ytck)
if j>0: ax.set_yticklabels([])
else: ax.set_yticklabels(ytck-[0.5,0.4,0.3,0.2,0.1,0.05][i])
#if i==1: plt.text(2,-1,FIG[3][2],size=8)
if i<5: ax.set_xticklabels([])
#else: plt.ylabel('subject %d'%(i+1))
#if i==0: plt.title(str(m[j]*2+2))
#
if i==0: plt.title(FIG[3][3]+str(j+1))
plt.subplot(6,4,5)
plt.text(-9,-0.5,FIG[3][1],rotation='vertical')
plt.subplot(6,4,22)
plt.text(-2,-1.2,FIG[3][2])
plt.subplots_adjust(wspace=-1)
plt.savefig(figpath+FIG[3][0]+'avg',dpi=DPI,bbox_inches='tight')
est=np.array(est)
print est.ndim, est
if est.ndim>2: est=np.squeeze(est)
print ndarray2latextable(est,decim=2)
if not MOVIE: return dat
lbl=[]
#for i in range(4):lbl.append([FIG[6][1][3]+str(i+1),20,32+i*72,-15])
for i in range(4):lbl.append([FIG[3][3]+str(i+1),20,32+i*72,FIG[3][4]])
#for i in range(6):lbl.append([str([500,400,300,200,100,50][i]),20,-10,30+i*72])
for i in range(6):lbl.append([str([500,400,300,200,100,50][i]),20,-10,FIG[3][5]+i*FIG[3][6]])
plotGifGrid(dat,fn=figpath+'buttonPressMean'+FMT,bcgclr=1,
text=lbl,plottime=True)
return dat
def plotBTpt(vpn=range(1,5),pcaEv=97):
from scipy.optimize import fmin
def fun(x,D=None,verbose=False):
nrlines=len(x)/3; p1=np.nan;s=0.15
if nrlines==1: p0,v0,s0=tuple(x)
elif nrlines==2: p0,v0,s0,p1,v1,s1=tuple(x)
else: raise ValueError
out=np.ones((P,T))
dist=np.abs(pm+v0*tm-p0)/np.sqrt(1+v0**2)/s0
out=np.maximum(1-np.power(dist,3),0)/float(nrlines)
if nrlines==2:
dist=np.abs(pm+v1*tm-p1)/np.sqrt(1+v1**2)/s1
out+=np.maximum(1-np.power(dist,3),0)/float(nrlines)
out/=out.sum()
if D is None: return out
fout=-np.corrcoef(D.flatten(),out.flatten())[0,1]# np.linalg.norm(D-out)**2
if verbose: print 'p0=%.2f, v=%.2f, p1=%.2f, s=%.2f, f=%f'%(p0,v,p1,s,fout)
return fout
#dat=plotBTmean()
T=68#dat[0][0].shape[-1]
P=64#dat[0][0].shape[0]
t=np.linspace(-0.8,0,T);p=np.linspace(-5,5,P)
tm=np.repeat(t[np.newaxis,:],P,axis=0)
pm=np.repeat(p[:,np.newaxis],T,axis=1)
rows=len(vpn)
#cols=len(dat[0])
fig=figure(size=3,aspect=0.35)
fig.tight_layout()
#m=[-251,-201,-151,-101,-51,-1]
bnds=[(1,None),(None,0),(None,None)]
est=[]
for i in range(-1,rows):
#for k in [1]:
#est.append([])
j=4# 200 ms
fn=inpath+'vp%03d/E%d/'%(vpn[max(0,i)],pcaEv)+'X/coeff.npy'
if i==0: pc=(_getPC(np.load(fn),0)+_getPC(np.load(fn),1))/2.
elif i==-1: pc=(_getPC(np.load(fn),0)-_getPC(np.load(fn),1)+1)/2.
else: pc=_getPC(np.load(fn),0)
if pc.mean()>=0.4: pc=1-pc
inc=[-2,2,0,0,0,0][i+1]
D= pc.T[:,(31+inc):(33+inc),:].mean(1)
#else: D=dat[i][j][31:33,:,:].mean(0)
D/=D.sum()
#print i,k,D.shape,D.sum()
subplot(1,5,i+2)
plt.pcolor(p,t,D.T,cmap='gray')
# below we set the initial guess
if vpn[i]==999: x0=np.array((3,-12,0.1,7,-12,0.1))
elif vpn[max(0,i)]==1:
if i==-1: x0=np.array((1,-12,0.3,-2,-12,0.1))
else:x0=np.array((3,-12,0.1,0,-12,0.1))
else: x0=np.array((3,-12,0.1,-2,-12,0.1))
xopt=fmin(func=fun,x0=x0,args=(D,False))
est.append(xopt.tolist())
plt.plot(xopt[0]-xopt[1]*t,t,'r',lw=1,alpha=0.4)
#if vpn[i]!=1:
plt.plot(xopt[3]-xopt[4]*t,t,'r',lw=1,alpha=0.4)
#else:est[-1].extend([np.nan]*3)
plt.grid(True,'both');
plt.xlim([p[0],p[-1]]);plt.ylim([t[0],t[-1]]);
ax=plt.gca();ax.set_axisbelow(False)
ax.set_xticks([-4,-2,0,2,4])
ax.set_yticks(np.linspace(-0.8,0,5))
if i in set((0,1,2,3)): ax.set_yticklabels([])
else:
ax.set_yticklabels(np.linspace(-1,-0.2,5))
plt.ylabel(FIG[3][1])
#if i==1: plt.text(2,-1,FIG[3][2],size=8)
if i==1: plt.xlabel(FIG[3][2])
#else: plt.ylabel('subject %d'%(i+1))
#if i==0: plt.title(str(m[j]*2+2))
plt.text(1,-0.75,FIG[3][3]+str(vpn[max(i,0)])+['a','b',''][min(i+1,2)],color='w')
plt.subplots_adjust(wspace=-1)
plt.savefig(figpath+FIG[3][0]+'%d'%pcaEv,
dpi=DPI,bbox_inches='tight')
est=np.array(est)
print est.ndim, est
if est.ndim>2: est=np.squeeze(est)
print ndarray2latextable(est,decim=2)