def main():
fn_list = [(np.dot,'dot product')]
sigmas = list(np.linspace(4, 6,3))
for sigma in sigmas:
new_fn = partial(gaussian_rbf, sigma=sigma)
infotext = "RBF sigma = " + str(sigma)
fn_list.append((new_fn,infotext))
polys = [3,5]
for poly in polys:
new_fn = partial(poly_kernel, p=poly)
infotext = "Poly p = " + str(poly)
fn_list.append((new_fn, infotext))
gsk = [.5, 1, 5]
for gauss in gsk:
kernel = b.gaussk(gauss)
eta = .1
r = 15
bhatta_k = BhattaVectorWrapper(kernel, eta, r).bhatta
infotext = "bhatta w/ gaussk {}".format(gauss)
fn_list.append((bhatta_k, infotext))
for (fn, descriptor) in fn_list:
print2("Kernel " + descriptor + ": Error rate: " + str(perceptron_kernel(fn)))
def main():
fn_list = [(np.dot, 'dot product')]
sigmas = list(np.linspace(4, 6, 3))
for sigma in sigmas:
new_fn = partial(gaussian_rbf, sigma=sigma)
infotext = "RBF sigma = " + str(sigma)
fn_list.append((new_fn, infotext))
polys = [3, 5]
for poly in polys:
new_fn = partial(poly_kernel, p=poly)
infotext = "Poly p = " + str(poly)
fn_list.append((new_fn, infotext))
gsk = [.5, 1, 5]
for gauss in gsk:
kernel = b.gaussk(gauss)
eta = .1
r = 15
bhatta_k = BhattaVectorWrapper(kernel, eta, r).bhatta
infotext = "bhatta w/ gaussk {}".format(gauss)
fn_list.append((bhatta_k, infotext))
for (fn, descriptor) in fn_list:
print2("Kernel " + descriptor + ": Error rate: " +
str(perceptron_kernel(fn)))
X2 = matrix(X2)
X1x = X1[:,0]
X1y = X1[:,1]
X2x = X2[:,0]
X2y = X2[:,1]
#plt.figure()
#nk_b = nk_bhatta(X1, X2, 0)
#kvals = "Non-kernelized: " + "{:.2f}\n".format(nk_b).lstrip('0')
table_vals = []
for eta in etas:
sig_vals = []
for sigma in sigmas:
if verbose: print "Evaluating {:s} for e={:.2f} s={:.2f}"\
.format(title_text, eta, sigma)
kappa = bhatta.eig_bhatta(X1, X2, bhatta.gaussk(sigma), eta, 25)
sig_vals.append(kappa)
table_vals.append(sig_vals)
plt.figure()
plt.plot(X1x, X1y, 'ro', X2x, X2y, 'bo')
plt.axis([xmin, xmax, ymin, ymax])
col_labels = [r'$\sigma = %.2f$' % sig for sig in sigmas]
row_labels = [r'$\eta = %.2f$' % eta for eta in etas]
#kval_x, kval_y = xmin + x_range * .8, ymin + y_range * .7
gen1_x, gen1_y = xmin + x_range * .02, ymin + y_range * .8
gen2_x, gen2_y = xmin + x_range * .02, ymin + y_range * .6
X2 = matrix(X2)
X1x = X1[:, 0]
X1y = X1[:, 1]
X2x = X2[:, 0]
X2y = X2[:, 1]
#plt.figure()
#nk_b = nk_bhatta(X1, X2, 0)
#kvals = "Non-kernelized: " + "{:.2f}\n".format(nk_b).lstrip('0')
table_vals = []
for eta in etas:
sig_vals = []
for sigma in sigmas:
if verbose: print "Evaluating {:s} for e={:.2f} s={:.2f}"\
.format(title_text, eta, sigma)
kappa = bhatta.eig_bhatta(X1, X2, bhatta.gaussk(sigma), eta, 25)
sig_vals.append(kappa)
table_vals.append(sig_vals)
plt.figure()
plt.plot(X1x, X1y, 'ro', X2x, X2y, 'bo')
plt.axis([xmin, xmax, ymin, ymax])
col_labels = [r'$\sigma = %.2f$' % sig for sig in sigmas]
row_labels = [r'$\eta = %.2f$' % eta for eta in etas]
#kval_x, kval_y = xmin + x_range * .8, ymin + y_range * .7
gen1_x, gen1_y = xmin + x_range * .02, ymin + y_range * .8
gen2_x, gen2_y = xmin + x_range * .02, ymin + y_range * .6
