m_gam2 = r(gamcall)
ranova = r('anova')
print r['summary'](m_gam2)
anovares = ranova(m_gam, m_gam2, test = "Chisq")
print anovares # ANOVA correctly detect a better fit if separated.
print anovares[4] # p- value
def predict(i):
r('m0.pred <- predict(m.gam2, newdata = nd%d, type = "link")' % i) # predict from m.gam object, using new data
r('m0.pred <- m0.pred - mean(m0.pred)') # centering at zero
#m_zero = r('m0.pred + mean(c(0, coef(m0.glm)))') # zero as anchor
m_zero = r('m0.pred - m0.pred[1]') # zero as anchor
return np.array(m_zero)
mzeros = np.zeros((len(svec), len(files)))
for i in range(len(files)):
mzeros[:,i] = predict(i)
##
plt.figure()
plt.plot(svec, m_zero, label="GAM all", linewidth=2)
for i in range(len(files)):
plt.plot(svec, mzeros[:,i], label="GAM %d" % i, linewidth=2)
mlds.plotscale(objs[i], observer="GLM %d" % i, marker='o', linewidth=0)
plt.legend(loc=2)
plt.show()
glms=[]
for f in files:
O = mlds.MLDSObject(f, boot= True, standardscale= st, save=True)
#O.parallel=True
#O.run()
O.load()
glms.append( O )
obsGAM = mlds_gam.MLDSGAMCompare(files, standardscale = st)
obsGAM.run()
pval = obsGAM.anovares[4]
print pval
l = plt.plot(obsGAM.stim, obsGAM.scales, linewidth=2)
for o, obs in enumerate(glms):
plotscale(obs, '$\gamma = %.2f$' % gammas[o], l[o].get_color(), linewidth=0, marker='o')
#plt.plot(obs.stim, obs.scale, linewidth=0, marker='o', markerfacecolor = l[o].get_color(), label='$\gamma = %.2f$' % gammas[o])
ax = setplotproperties(ax)
plt.legend(loc=2, frameon=False)
plt.title('Example using power fn', {'fontsize': 10} )
fig.savefig('example.pdf')
plt.show()
######
# 3. Now we simulate the observer performing the method of triads.
# The simulated results are stored in the file fname
# (Reduce nblocks to 1 if you just want to see the result format.)
fname = mlds.simulateobserver(fn, stim, nblocks=15)
#######
# 4. Finally, we want to estimate the scale from the simulated data.
obs = mlds.MLDSObject(fname, boot=True, standardscale=False, verbose=True)
obs.load() # this takes a while as bootstrap is done
obs.printinfo()
# we can plot the scale
# the shape of the scale should coincide with a power function with exponent 2
mlds.plotscale(obs)
plt.xlim([-0.1, 1.1])
plt.show()
# and the noise estimated by MLDS should be the double of the noise
# introduced at the sensory level in the simulation
print("noise estimate from MLDS: %.3f" % (1 / obs.mns[-1]))
print("which must be the double of the sensory noise %.3f" % fn.sigmamax)
# finally, GoF measures should be OK, as we are simulating an observer
# that actually performs the decision model assumed by MLDS
obs.rundiagnostics()
print("GoF measures:")
print('AIC: %f, DAF: %f' % (obs.AIC, obs.DAF))
print('p-val: %f' % obs.prob)
###### # 3. Now we simulate the observer performing the method of triads. # The simulated results are stored in the file fname fname = mlds.simulateobserver(fn, stim, nblocks=15) ####### # 4. Finally, we want to estimate the scale from the simulated data. obs = mlds.MLDSObject(fname, boot=True, standardscale=False) obs.load() # this takes a while as bootstrap is done obs.printinfo # we can plot the scale # the shape of the scale should coincide with a power function with # exponent 2 mlds.plotscale(obs) plt.xlim([-0.1, 1.1]) plt.show() # and the noise estimated by MLDS should be the double of the noise # introduced at the sensory level in the simulation print "noise estimate from MLDS: %.3f" % (1/obs.mns[-1]) print "which must be the double of the sensory noise %.3f" % fn.sigmamax # finally, GoF measures should be OK, as we are simulating an observer # that actually performs the decision model assumed by MLDS # obs.rundiagnostics() # print "GoF measures:" # print 'AIC: %f, DAF: %f' % (obs.AIC, obs.DAF) # print 'p-val: %f' % obs.prob