def _testTreeAccuracy(self):
RF1 = RandomForest(ntrees=1, m=4, ndata=2, pRetrain=0.2)
RF10 = RandomForest(ntrees=10, m=4, ndata=2, pRetrain=0.2)
RF100 = RandomForest(ntrees=10, m=4, ndata=2, pRetrain=0.2)
tf = Branin()
X = lhcSample(tf.bounds, 20, seed=0)
Y = [tf.f(x) for x in X]
RF1.addData(X, Y)
RF10.addData(X, Y)
RF100.addData(X, Y)
rmse1 = 0.0
rmse10 = 0.0
rmse100 = 0.0
nsamp = 1000
for testx in lhcSample(tf.bounds, nsamp, seed=1):
testy = tf.f(testx)
rmse1 += (RF1.mu(testx) - testy) ** 2
rmse10 += (RF10.mu(testx) - testy) ** 2
rmse100 += (RF100.mu(testx) - testy) ** 2
# this isn't consistent, since random forests are, you know, random
print "RMSE 1 = %.4f" % (rmse1 / nsamp)
print "RMSE 10 = %.4f" % (rmse10 / nsamp)
print "RMSE 100 = %.4f" % (rmse100 / nsamp)
self.failUnless(rmse1 > rmse100)
def _testTreeAccuracy(self):
RF1 = RandomForest(ntrees=1, m=4, ndata=2, pRetrain=.2)
RF10 = RandomForest(ntrees=10, m=4, ndata=2, pRetrain=.2)
RF100 = RandomForest(ntrees=10, m=4, ndata=2, pRetrain=.2)
tf = Branin()
X = lhcSample(tf.bounds, 20, seed=0)
Y = [tf.f(x) for x in X]
RF1.addData(X, Y)
RF10.addData(X, Y)
RF100.addData(X, Y)
rmse1 = 0.0
rmse10 = 0.0
rmse100 = 0.0
nsamp = 1000
for testx in lhcSample(tf.bounds, nsamp, seed=1):
testy = tf.f(testx)
rmse1 += (RF1.mu(testx) - testy)**2
rmse10 += (RF10.mu(testx) - testy)**2
rmse100 += (RF100.mu(testx) - testy)**2
# this isn't consistent, since random forests are, you know, random
print 'RMSE 1 = %.4f' % (rmse1 / nsamp)
print 'RMSE 10 = %.4f' % (rmse10 / nsamp)
print 'RMSE 100 = %.4f' % (rmse100 / nsamp)
self.failUnless(rmse1 > rmse100)
def testOneTree(self):
forest = RandomForest(ntrees=1, m=4, ndata=2, pRetrain=.2)
self.failUnlessEqual(forest.ntrees, 1)
self.failUnlessEqual(forest.m, 4)
self.failUnlessEqual(forest.ndata, 2)
self.failUnlessEqual(forest.pRetrain, .2)
tf = Branin()
X = lhcSample(tf.bounds, 20, seed=0)
Y = [tf.f(x) for x in X]
forest.addData(X, Y)
self.failUnlessEqual(len(forest.forest), 1)
# checkTree(forest.forest[0])
# maximizeEI(forest, tf.bounds)
# print forest.forest[0]
if False:
figure(1, figsize=(5, 10))
c0 = [(i / 100.) * (tf.bounds[0][1] - tf.bounds[0][0]) +
tf.bounds[0][0] for i in xrange(101)]
c1 = [(i / 100.) * (tf.bounds[1][1] - tf.bounds[1][0]) +
tf.bounds[1][0] for i in xrange(101)]
ax = subplot(121)
mu = array([[forest.mu(array([i, j])) for i in c0] for j in c1])
cs = ax.contourf(c0, c1, mu, 50)
colorbar(cs)
ax.plot([x[0] for x in X], [x[1] for x in X], 'ro', alpha=.2)
ax.set_xbound(tf.bounds[0][0], tf.bounds[0][1])
ax.set_ybound(tf.bounds[1][0], tf.bounds[1][1])
ax.set_title(r'$\mu$')
ax = subplot(122)
mu = array([[forest.sigma2(array([i, j])) for i in c0]
for j in c1])
cs = ax.contourf(c0, c1, mu, 50)
colorbar(cs)
ax.plot([x[0] for x in X], [x[1] for x in X], 'ro', alpha=.2)
ax.set_xbound(tf.bounds[0][0], tf.bounds[0][1])
ax.set_ybound(tf.bounds[1][0], tf.bounds[1][1])
ax.set_title(r'$\sigma^2$')
show()
def testOneTree(self):
forest = RandomForest(ntrees=1, m=4, ndata=2, pRetrain=0.2)
self.failUnlessEqual(forest.ntrees, 1)
self.failUnlessEqual(forest.m, 4)
self.failUnlessEqual(forest.ndata, 2)
self.failUnlessEqual(forest.pRetrain, 0.2)
tf = Branin()
X = lhcSample(tf.bounds, 20, seed=0)
Y = [tf.f(x) for x in X]
forest.addData(X, Y)
self.failUnlessEqual(len(forest.forest), 1)
# checkTree(forest.forest[0])
# maximizeEI(forest, tf.bounds)
# print forest.forest[0]
if False:
figure(1, figsize=(5, 10))
c0 = [(i / 100.0) * (tf.bounds[0][1] - tf.bounds[0][0]) + tf.bounds[0][0] for i in xrange(101)]
c1 = [(i / 100.0) * (tf.bounds[1][1] - tf.bounds[1][0]) + tf.bounds[1][0] for i in xrange(101)]
ax = subplot(121)
mu = array([[forest.mu(array([i, j])) for i in c0] for j in c1])
cs = ax.contourf(c0, c1, mu, 50)
colorbar(cs)
ax.plot([x[0] for x in X], [x[1] for x in X], "ro", alpha=0.2)
ax.set_xbound(tf.bounds[0][0], tf.bounds[0][1])
ax.set_ybound(tf.bounds[1][0], tf.bounds[1][1])
ax.set_title(r"$\mu$")
ax = subplot(122)
mu = array([[forest.sigma2(array([i, j])) for i in c0] for j in c1])
cs = ax.contourf(c0, c1, mu, 50)
colorbar(cs)
ax.plot([x[0] for x in X], [x[1] for x in X], "ro", alpha=0.2)
ax.set_xbound(tf.bounds[0][0], tf.bounds[0][1])
ax.set_ybound(tf.bounds[1][0], tf.bounds[1][1])
ax.set_title(r"$\sigma^2$")
show()
def testForestUCB(self):
RF = RandomForest(ntrees=2)
tf = Branin()
X = lhcSample(tf.bounds, 20, seed=0)
Y = [tf.f(x) for x in X]
RF.addData(X, Y)
ucbf = UCB(RF, len(tf.bounds))
dopt, doptx = direct(ucbf.negf, tf.bounds, maxiter=10)
copt, coptx = cdirect(ucbf.negf, tf.bounds, maxiter=10)
mopt, moptx = maximizeUCB(RF, tf.bounds, maxiter=10)
self.failUnlessAlmostEqual(dopt, copt, 4)
self.failUnlessAlmostEqual(-dopt, mopt, 4)
self.failUnlessAlmostEqual(-copt, mopt, 4)
self.failUnless(sum(abs(doptx - coptx)) < 0.01)
self.failUnless(sum(abs(moptx - coptx)) < 0.01)
self.failUnless(sum(abs(moptx - doptx)) < 0.01)
def testForestUCB(self):
RF = RandomForest(ntrees=2)
tf = Branin()
X = lhcSample(tf.bounds, 20, seed=0)
Y = [tf.f(x) for x in X]
RF.addData(X, Y)
ucbf = UCB(RF, len(tf.bounds))
dopt, doptx = direct(ucbf.negf, tf.bounds, maxiter=10)
copt, coptx = cdirect(ucbf.negf, tf.bounds, maxiter=10)
mopt, moptx = maximizeUCB(RF, tf.bounds, maxiter=10)
self.failUnlessAlmostEqual(dopt, copt, 4)
self.failUnlessAlmostEqual(-dopt, mopt, 4)
self.failUnlessAlmostEqual(-copt, mopt, 4)
self.failUnless(sum(abs(doptx - coptx)) < .01)
self.failUnless(sum(abs(moptx - coptx)) < .01)
self.failUnless(sum(abs(moptx - doptx)) < .01)
def testForestPI(self):
RF = RandomForest(ntrees=2)
tf = Branin()
X = lhcSample(tf.bounds, 20, seed=0)
Y = [tf.f(x) for x in X]
RF.addData(X, Y)
mu, sigma = RF.posterior(ones(len(tf.bounds)) * 0.4)
print "[python] = 0.4 x 2, mu =", mu, " sigma =", sigma
pif1 = PI(RF)
dopt1, doptx1 = direct(pif1.negf, tf.bounds, maxiter=10)
copt1, coptx1 = cdirect(pif1.negf, tf.bounds, maxiter=10)
mopt1, moptx1 = maximizePI(RF, tf.bounds, maxiter=10)
self.failUnlessAlmostEqual(dopt1, copt1, 4)
self.failUnlessAlmostEqual(-dopt1, mopt1, 4)
self.failUnlessAlmostEqual(-copt1, mopt1, 4)
self.failUnless(sum(abs(doptx1 - coptx1)) < 0.01)
self.failUnless(sum(abs(moptx1 - coptx1)) < 0.01)
self.failUnless(sum(abs(moptx1 - doptx1)) < 0.01)
pif2 = PI(RF, xi=0.5)
dopt2, doptx2 = direct(pif2.negf, tf.bounds, maxiter=10)
copt2, coptx2 = cdirect(pif2.negf, tf.bounds, maxiter=10)
mopt2, moptx2 = maximizePI(RF, tf.bounds, xi=0.5, maxiter=10)
self.failUnlessAlmostEqual(dopt2, copt2, 4)
self.failUnlessAlmostEqual(-dopt2, mopt2, 4)
self.failUnlessAlmostEqual(-copt2, mopt2, 4)
self.failUnless(sum(abs(doptx2 - coptx2)) < 0.01)
self.failUnless(sum(abs(moptx2 - coptx2)) < 0.01)
self.failUnless(sum(abs(moptx2 - doptx2)) < 0.01)
self.failIfAlmostEqual(dopt1, dopt2, 4)
self.failIfAlmostEqual(copt1, copt2, 4)
self.failIfAlmostEqual(mopt1, mopt2, 4)
def testForestPI(self):
RF = RandomForest(ntrees=2)
tf = Branin()
X = lhcSample(tf.bounds, 20, seed=0)
Y = [tf.f(x) for x in X]
RF.addData(X, Y)
mu, sigma = RF.posterior(ones(len(tf.bounds)) * .4)
print '[python] = 0.4 x 2, mu =', mu, ' sigma =', sigma
pif1 = PI(RF)
dopt1, doptx1 = direct(pif1.negf, tf.bounds, maxiter=10)
copt1, coptx1 = cdirect(pif1.negf, tf.bounds, maxiter=10)
mopt1, moptx1 = maximizePI(RF, tf.bounds, maxiter=10)
self.failUnlessAlmostEqual(dopt1, copt1, 4)
self.failUnlessAlmostEqual(-dopt1, mopt1, 4)
self.failUnlessAlmostEqual(-copt1, mopt1, 4)
self.failUnless(sum(abs(doptx1 - coptx1)) < .01)
self.failUnless(sum(abs(moptx1 - coptx1)) < .01)
self.failUnless(sum(abs(moptx1 - doptx1)) < .01)
pif2 = PI(RF, xi=0.5)
dopt2, doptx2 = direct(pif2.negf, tf.bounds, maxiter=10)
copt2, coptx2 = cdirect(pif2.negf, tf.bounds, maxiter=10)
mopt2, moptx2 = maximizePI(RF, tf.bounds, xi=0.5, maxiter=10)
self.failUnlessAlmostEqual(dopt2, copt2, 4)
self.failUnlessAlmostEqual(-dopt2, mopt2, 4)
self.failUnlessAlmostEqual(-copt2, mopt2, 4)
self.failUnless(sum(abs(doptx2 - coptx2)) < .01)
self.failUnless(sum(abs(moptx2 - coptx2)) < .01)
self.failUnless(sum(abs(moptx2 - doptx2)) < .01)
self.failIfAlmostEqual(dopt1, dopt2, 4)
self.failIfAlmostEqual(copt1, copt2, 4)
self.failIfAlmostEqual(mopt1, mopt2, 4)
