bw = experiment.m_vb.likelihood.binsize
plt.plot(X,np.mean(np.exp(experiment.fsamples_vb)*bw,axis=1) ,'b' ,label='VB+Gaussian')
plt.plot(X,np.mean(np.exp(experiment.fsamples_mc)*bw,axis=1) ,'r' ,label='VB+MCMC')
plt.legend()
plt.plot(X,np.percentile(np.exp(experiment.fsamples_vb)*bw, 5,axis=1), 'b-')
plt.plot(X,np.percentile(np.exp(experiment.fsamples_vb)*bw,95,axis=1), 'b-')
plt.plot(X,np.percentile(np.exp(experiment.fsamples_mc)*bw, 5,axis=1), 'r-')
plt.plot(X,np.percentile(np.exp(experiment.fsamples_mc)*bw,95,axis=1), 'r-')
#load and plot raw data
X = load_coal()
plt.plot(X, X*0, 'k|')
plt.xlabel('time (years)')
plt.ylabel('rate')
plt.ylim(-.05, 1)
plt.xlim(X.min(), X.max())
save_tikz('coal_rates.tikz',figurewidth='\\figurewidth', figureheight = '\\figureheight')
plt.figure()
#trans = GPy.core.parameterization.transformations.Logexp()
trans = GPy.core.parameterization.transformations.Exponent()
plt.hist(trans.f(experiment.samples[:,0]), 100, normed=True)
plt.xlabel('signal varaince')
#save_tikz('coal_variance.tikz',figurewidth='\\figurewidth', figureheight = '\\figureheight')
np.savetxt('coal_var_samples',trans.f(experiment.samples[:,0]))
plt.figure()
plt.hist(trans.f(experiment.samples[:,1]), 100, normed=True)
plt.xlabel('lengthscale')
#save_tikz('coal_lengthscale.tikz',figurewidth='\\figurewidth', figureheight = '\\figureheight')
np.savetxt('coal_ls_samples',trans.f(experiment.samples[:,1]))
#plota scatter of variance, ls
samples[pointIndex,:] = means[currentPointCategory] + np.dot(Ls[currentPointCategory],rng.randn(2) )
categories[pointIndex] = currentPointCategory
return samples, categories
if __name__ == '__main__':
nTrainingSamples = 750
nTestingSamples = 750
rng = np.random.RandomState( 1 )
X_train, Y_train = getSamples( nTrainingSamples, rng )
X_test, Y_test = getSamples( nTrainingSamples, rng )
from matplotlib import pylab as plt
RTrain = np.array( [ X_train[index,:] for index in range(X_train.shape[0] ) if Y_train[index]== 0 ] )
GTrain = np.array( [ X_train[index,:] for index in range(X_train.shape[0] ) if Y_train[index]== 1 ] )
BTrain = np.array( [ X_train[index,:] for index in range(X_train.shape[0] ) if Y_train[index]== 2 ] )
plt.plot( RTrain[:,0], RTrain[:,1] ,'ro')
plt.plot( GTrain[:,0], GTrain[:,1] , 'bo')
plt.plot( BTrain[:,0], BTrain[:,1] , 'go')
plt.xlim( [-6.,6.] )
plt.ylim( [-6.,6.] )
data = {'X_train':X_train,'Y_train':Y_train,'X_test':X_test,'Y_test':Y_test}
cPickle.dump( data , open('data','w') )
from matplotlib2tikz import save as save_tikz
save_tikz('simple_data.tikz')
m_mc.likelihood.delta.fix(1e-3)
m_vb.Z.fix()
m_vb.kern.fix()
#m_vb.optimize('bfgs', max_iters=vb_frozen_iters, messages=True)
#m_vb.Z.unfix()
#m_vb.kern.constrain_positive()
#m_vb.kern.white.fix(1e-3) # to keep the same as mcmc
#m_vb.optimize('bfgs', max_iters=vb_max_iters, messages=True)
m_vb = GPy.load('m_vb')
drawPredictionSpace(m_vb)
from matplotlib2tikz import save as save_tikz
save_tikz('simple_vb.tikz')
#m_mc.kern.rbf[:] = m_vb.kern.rbf[:]*1
m_mc.Z[:] = m_vb.Z[:] * 1
m_mc.Z.fix()
L = GPy.util.choleskies.flat_to_triang(m_vb.q_u_chol)
U = np.vstack(
[np.dot(L[i, :, :], np.random.randn(m_mc.V.shape[0])) for i in range(3)]).T
U = U + m_vb.q_u_mean
K = m_mc.kern.K(m_mc.Z)
L = GPy.util.linalg.jitchol(K)
m_mc.V[:] = GPy.util.linalg.dtrtrs(L, U)[0]
#samples mcmc
def f(x):
epochs_test_loss=epochs_test_loss)
plt.figure(0)
plt.xlabel("Iteration")
plt.ylabel("Loss")
plt.plot(train_batches_loss, "g-", label="Loss", alpha=0.2)
plt.plot(moving_average(train_batches_loss,
moving_average_window_size,
use_fraction=True),
"g-",
label="Loss MA")
plt.legend()
plt.tight_layout()
plt.savefig(os.path.join(path, "train_batches_loss.png"))
save_tikz(os.path.join(path, "train_batches_loss.tikz"),
figureheight="\\figureheight",
figurewidth="\\figurewidth")
plt.figure(1)
plt.xlabel("Iteration")
plt.ylabel("Accuracy")
plt.plot(train_batches_accuracy, "g-", label="Accuracy", alpha=0.2)
plt.plot(moving_average(train_batches_accuracy,
moving_average_window_size,
use_fraction=True),
"g-",
label="Accuracy MA")
plt.legend()
plt.tight_layout()
plt.savefig(os.path.join(path, "train_batches_accuracy.png"))
save_tikz(os.path.join(path, "train_batches_accuracy.tikz"),
def output_current_plot(fileName):
save_tikz(fileName,
figurewidth='\\figurewidth',
figureheight='\\figureheight')
'b-')
plt.plot(X, np.percentile(np.exp(experiment.fsamples_vb) * bw, 95, axis=1),
'b-')
plt.plot(X, np.percentile(np.exp(experiment.fsamples_mc) * bw, 5, axis=1),
'r-')
plt.plot(X, np.percentile(np.exp(experiment.fsamples_mc) * bw, 95, axis=1),
'r-')
#load and plot raw data
X = load_coal()
plt.plot(X, X * 0, 'k|')
plt.xlabel('time (years)')
plt.ylabel('rate')
plt.ylim(-.05, 1)
plt.xlim(X.min(), X.max())
save_tikz('coal_rates.tikz',
figurewidth='\\figurewidth',
figureheight='\\figureheight')
plt.figure()
#trans = GPy.core.parameterization.transformations.Logexp()
trans = GPy.core.parameterization.transformations.Exponent()
plt.hist(trans.f(experiment.samples[:, 0]), 100, normed=True)
plt.xlabel('signal varaince')
#save_tikz('coal_variance.tikz',figurewidth='\\figurewidth', figureheight = '\\figureheight')
np.savetxt('coal_var_samples', trans.f(experiment.samples[:, 0]))
plt.figure()
plt.hist(trans.f(experiment.samples[:, 1]), 100, normed=True)
plt.xlabel('lengthscale')
#save_tikz('coal_lengthscale.tikz',figurewidth='\\figurewidth', figureheight = '\\figureheight')
np.savetxt('coal_ls_samples', trans.f(experiment.samples[:, 1]))
size=12,
ha="center",
transform=ax[i].transAxes)
ax[i].xaxis.set_major_locator(plt.NullLocator())
ax[i].yaxis.set_major_locator(plt.NullLocator())
ax[i].imshow(groups[i][0][:, :, [2, 1, 0]],
cmap="gray",
vmin=0,
vmax=1)
plt.tight_layout()
if arg_write:
plt.savefig(
os.path.join(plot_path,
"_".join([obj_name, "grouping.png"])))
save_tikz(os.path.join(
plot_path, "_".join([obj_name, "grouping.tikz"])),
figureheight="\\figureheight",
figurewidth="\\figurewidth")
if not arg_groups:
plt.close()
#plots saliency maps of each view
if arg_saliency or arg_write:
if scores.size and group_ids.size and group_weights.size:
groups = sorted(zip(views, scores, group_ids, saliency),
key=lambda x: (x[2], x[1]))
saliency = [s[3] for s in groups]
fig, ax = plt.subplots(rows_in_plot,
int(arg_views / rows_in_plot),
figsize=(6.2, 2),
dpi=300)
plt.subplots_adjust()
from matplotlib import pylab as plt
RTrain = np.array([
X_train[index, :] for index in range(X_train.shape[0])
if Y_train[index] == 0
])
GTrain = np.array([
X_train[index, :] for index in range(X_train.shape[0])
if Y_train[index] == 1
])
BTrain = np.array([
X_train[index, :] for index in range(X_train.shape[0])
if Y_train[index] == 2
])
plt.plot(RTrain[:, 0], RTrain[:, 1], 'ro')
plt.plot(GTrain[:, 0], GTrain[:, 1], 'bo')
plt.plot(BTrain[:, 0], BTrain[:, 1], 'go')
plt.xlim([-6., 6.])
plt.ylim([-6., 6.])
data = {
'X_train': X_train,
'Y_train': Y_train,
'X_test': X_test,
'Y_test': Y_test
}
cPickle.dump(data, open('data', 'w'))
from matplotlib2tikz import save as save_tikz
save_tikz('simple_data.tikz')
m_vb.likelihood.delta.fix(1e-3)
m_mc.likelihood.delta.fix(1e-3)
m_vb.Z.fix()
m_vb.kern.fix()
#m_vb.optimize('bfgs', max_iters=vb_frozen_iters, messages=True)
#m_vb.Z.unfix()
#m_vb.kern.constrain_positive()
#m_vb.kern.white.fix(1e-3) # to keep the same as mcmc
#m_vb.optimize('bfgs', max_iters=vb_max_iters, messages=True)
m_vb = GPy.load('m_vb')
drawPredictionSpace( m_vb )
from matplotlib2tikz import save as save_tikz
save_tikz('simple_vb.tikz')
#m_mc.kern.rbf[:] = m_vb.kern.rbf[:]*1
m_mc.Z[:] = m_vb.Z[:]*1
m_mc.Z.fix()
L = GPy.util.choleskies.flat_to_triang(m_vb.q_u_chol)
U = np.vstack([np.dot(L[i,:,:], np.random.randn(m_mc.V.shape[0])) for i in range(3)]).T
U = U + m_vb.q_u_mean
K = m_mc.kern.K(m_mc.Z)
L = GPy.util.linalg.jitchol(K)
m_mc.V[:] = GPy.util.linalg.dtrtrs(L, U)[0]
#samples mcmc
def f(x):
return (-a for a in m_mc._objective_grads(x))
#m_mc.kern.fix()
