def optimize(self,obj=np.max,index=0):
c3_increment = self.coef_3_step_size*self.coef_3_multiplier
c2_increment = self.coef_2_step_size*self.coef_2_multiplier
c3_vec = np.arange(self.coef_3_min,self.coef_3_max+c3_increment,c3_increment)
c2_vec = np.arange(self.coef_2_min,self.coef_2_max+c2_increment,c2_increment)
scores = np.ones((len(c3_vec),len(c2_vec)))*obj(self.bscan)
winning_score = -np.inf
c3_winner = self.dispersion_coefs[0]
c2_winner = self.dispersion_coefs[1]
for idx3,c3 in enumerate(c3_vec):
for idx2,c2 in enumerate(c2_vec):
coefs = [c3,c2,0.0,0.0]
try:
test = self.proc_cache[(index,c3,c2)]
except KeyError as ke:
test = np.abs(process(self.raw_vol[index,:,:],self.k_in,self.k_out,coefs)[Z_CUTON:Z_CUTOFF,X_START:]).T
self.proc_cache[(index,c3,c2)] = test
score = obj(test)
scores[idx3,idx2] = score
if score>winning_score:
c3_winner = c3
c2_winner = c2
plt.imshow(scores,interpolation='none')
plt.pause(.1)
plt.colorbar()
self.coef_3_text.setValue(c3_winner/self.coef_3_multiplier)
self.coef_2_text.setValue(c2_winner/self.coef_2_multiplier)
self.change_coefs()
self.show_bscan()
def show_bscan(self,index=None):
if index is None:
index = self.index
try:
self.bscan = self.proc_cache[(index,self.dispersion_coefs[0],self.dispersion_coefs[1])]
except KeyError as ke:
self.bscan = np.abs(process(self.raw_vol[index,:,:],self.k_in,self.k_out,self.dispersion_coefs)[Z_CUTON:Z_CUTOFF,X_START:]).T
self.proc_cache[(index,self.dispersion_coefs[0],self.dispersion_coefs[1])] = self.bscan
self.compute_stats()
self.set_coef_labels()
self.canvas.setImage(self.bscan)
def __init__(self, unp_list, auto=False):
self.h5_list = []
self.coefs = []
self.k_ins = []
self.k_outs = []
self.n_slows = []
winners = []
for u in unp_list:
h5fn = u.replace(".unp", "") + ".hdf5"
h5 = H5(h5fn)
self.h5_list.append(h5)
self.k_ins.append(h5["k_in"][:])
self.k_outs.append(h5["k_out"][:])
self.coefs.append(h5["dispersion/coefficients"][:])
self.n_slows.append(h5["config/n_slow"].value)
for h5, kin, kout, n_slow in zip(self.h5_list, self.k_ins, self.k_outs, self.n_slows):
scores = []
for idx, coef in enumerate(self.coefs):
f = h5["raw_data"][0, int(n_slow / 2), 5:-5:, :]
p = np.abs(process(f, kin, kout, coef))
profile = np.mean(p, axis=1)
pmax = np.max(profile)
pmin = np.min(profile)
pcontrast = (pmax - pmin) / (pmax + pmin)
pgrad = np.max(np.abs(np.diff(profile)))
c = 2
score = pgrad * (pmax + c * pcontrast)
scores.append(score)
if not auto:
print "%d\t%0.1f\t%0.3f\t%0.1f\t%0.1f" % (idx + 1, pmax, pcontrast, pgrad, score)
plt.figure()
plt.subplot(1, 2, 1)
clim = np.percentile(p, (5, 99.9))
plt.imshow(p, interpolation="none", aspect="auto", clim=clim, cmap="gray")
plt.subplot(1, 2, 2)
plt.plot(profile)
if not auto:
plt.show()
winner = int(raw_input("Choose best figure number: ")) - 1
else:
# print 'scores :',scores
winner = np.argmax(scores)
print "winner: %d" % winner, self.coefs[winner]
winners.append(winner)
print "winners:", winners
mfw = max(set(winners), key=winners.count)
print "most frequent winner: ", mfw
def optimize2(self,obj=None,index=None):
if obj is None:
obj = self.gradient
if index is None:
index = self.index
c3_increment = self.coef_3_step_size*self.coef_3_multiplier
c2_increment = self.coef_2_step_size*self.coef_2_multiplier
#c3_vec = np.arange(self.coef_3_min,self.coef_3_max+c3_increment,c3_increment)
#c2_vec = np.arange(self.coef_2_min,self.coef_2_max+c2_increment,c2_increment)
c3_vec = np.arange(self.dispersion_coefs[0]-c3_increment*N_STEPS,self.dispersion_coefs[0]+c3_increment*(N_STEPS+1),c3_increment)
c2_vec = np.arange(self.dispersion_coefs[1]-c2_increment*N_STEPS,self.dispersion_coefs[1]+c2_increment*(N_STEPS+1),c2_increment)
c3_max = np.max(c3_vec)
c2_max = np.max(c2_vec)
c3_min = np.min(c3_vec)
c2_min = np.min(c2_vec)
c3 = self.dispersion_coefs[0]
c2 = self.dispersion_coefs[1]
scores = np.ones((len(c3_vec),len(c2_vec)))*obj(self.bscan)
print c3_min,c3,c3_max
print c2_min,c2,c2_max
while c3_min<=c3<=c3_max and c2_min<=c2<=c2_max:
print c3_min<=c3<=c3_max
print c2_min<=c2<=c2_max
best_score = -np.inf
for d3 in range(-1,2):
for d2 in range(-1,2):
coefs = [c3+d3*c3_increment,c2+d2*c2_increment,0.0,0.0]
try:
test = self.proc_cache[(index,c3,c2)]
except KeyError as ke:
test = np.abs(process(self.raw_vol[index,:,:],self.k_in,self.k_out,coefs)[Z_CUTON:Z_CUTOFF,X_START:]).T
self.proc_cache[(index,c3,c2)] = test
score = obj(test)
if score>best_score:
best_d3 = d3
best_d2 = d2
best_score = score
c3 = c3+best_d3*c3_increment
c2 = c2+best_d2*c2_increment
i3 = np.argmin(np.abs(c3-c3_vec))
i2 = np.argmin(np.abs(c2-c2_vec))
scores[i3,i2] = best_score
self.canvas.setImage(scores)
if best_d3==0 and best_d2==0:
break
c3_winner = c3
c2_winner = c2
self.coef_3_text.setValue(c3_winner/self.coef_3_multiplier)
self.coef_2_text.setValue(c2_winner/self.coef_2_multiplier)
self.change_coefs()
self.show_bscan()
