apunkt/GPy
1
0
Fork 0
mirror of https://github.com/SheffieldML/GPy.git synced 2026-05-13 22:12:38 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

some minor example modifications and cgd adjustments

Browse source
This commit is contained in:
Max Zwiessele 2013-05-03 14:22:18 +01:00
parent 9229100af7
commit 5321bfc8c9
4 changed files with 33 additions and 31 deletions
Download patch file Download diff file Expand all files Collapse all files

14
GPy/examples/dimensionality_reduction.py
View file

@ -176,13 +176,12 @@ def bgplvm_simulation_matlab_compare():
Y = sim_data['Y'] Y = sim_data['Y']
S = sim_data['S'] S = sim_data['S']
mu = sim_data['mu'] mu = sim_data['mu']
M, [_, Q] = 30, mu.shape M, [_, Q] = 20, mu.shape
Q = 2
from GPy.models import mrd from GPy.models import mrd
from GPy import kern from GPy import kern
reload(mrd); reload(kern) reload(mrd); reload(kern)
#k = kern.rbf(Q, ARD=True) + kern.bias(Q, np.exp(-2)) + kern.white(Q, np.exp(-2)) # k = kern.rbf(Q, ARD=True) + kern.bias(Q, np.exp(-2)) + kern.white(Q, np.exp(-2))
k = kern.linear(Q, ARD=True) + kern.bias(Q, np.exp(-2)) + kern.white(Q, np.exp(-2)) k = kern.linear(Q, ARD=True) + kern.bias(Q, np.exp(-2)) + kern.white(Q, np.exp(-2))
m = Bayesian_GPLVM(Y, Q, init="PCA", M=M, kernel=k, m = Bayesian_GPLVM(Y, Q, init="PCA", M=M, kernel=k,
# X=mu, # X=mu,
@ -191,14 +190,15 @@ def bgplvm_simulation_matlab_compare():
m.ensure_default_constraints() m.ensure_default_constraints()
m.auto_scale_factor = True m.auto_scale_factor = True
m['noise'] = Y.var() / 100. m['noise'] = Y.var() / 100.
m['linear_variance'] = .01
lscstr = '{}'.format(k.parts[0].name) # lscstr = '{}'.format(k.parts[0].name)
# m[lscstr] = .01 # m[lscstr] = .01
m.unconstrain(lscstr); m.constrain_fixed(lscstr, 10) # m.unconstrain(lscstr); m.constrain_fixed(lscstr, 10)
lscstr = 'X_variance' # lscstr = 'X_variance'
# m[lscstr] = .01 # m[lscstr] = .01
m.unconstrain(lscstr); m.constrain_fixed(lscstr, .1) # m.unconstrain(lscstr); m.constrain_fixed(lscstr, .1)
# cstr = 'white' # cstr = 'white'
# m.unconstrain(cstr); m.constrain_bounded(cstr, .01, 1.) # m.unconstrain(cstr); m.constrain_bounded(cstr, .01, 1.)

28
GPy/inference/conjugate_gradient_descent.py
View file

@ -166,25 +166,26 @@ class Async_Optimize(object):
except Empty: except Empty:
pass pass
def fmin_async(self, f, df, x0, callback, update_rule=FletcherReeves, def opt_async(self, f, df, x0, callback, update_rule=FletcherReeves,
messages=0, maxiter=5e3, max_f_eval=15e3, gtol=1e-6, messages=0, maxiter=5e3, max_f_eval=15e3, gtol=1e-6,
report_every=10, *args, **kwargs): report_every=10, *args, **kwargs):
self.runsignal.set() self.runsignal.set()
outqueue = Queue() outqueue = Queue()
c = None
if callback: if callback:
self.callback = callback self.callback = callback
c = Thread(target=self.async_callback_collect, args=(outqueue,)) c = Thread(target=self.async_callback_collect, args=(outqueue,))
c.start() c.start()
p = _CGDAsync(f, df, x0, update_rule, self.runsignal, self.SENTINEL, p = _CGDAsync(f, df, x0, update_rule, self.runsignal, self.SENTINEL,
report_every=report_every, messages=messages, maxiter=maxiter, report_every=report_every, messages=messages, maxiter=maxiter,
max_f_eval=max_f_eval, gtol=gtol, outqueue=outqueue, *args, **kwargs) max_f_eval=max_f_eval, gtol=gtol, outqueue=outqueue, *args, **kwargs)
p.run() p.run()
return p, c return p, c
def fmin(self, f, df, x0, callback=None, update_rule=FletcherReeves, def opt(self, f, df, x0, callback=None, update_rule=FletcherReeves,
messages=0, maxiter=5e3, max_f_eval=15e3, gtol=1e-6, messages=0, maxiter=5e3, max_f_eval=15e3, gtol=1e-6,
report_every=10, *args, **kwargs): report_every=10, *args, **kwargs):
p, c = self.fmin_async(f, df, x0, callback, update_rule, messages, p, c = self.opt_async(f, df, x0, callback, update_rule, messages,
maxiter, max_f_eval, gtol, maxiter, max_f_eval, gtol,
report_every, *args, **kwargs) report_every, *args, **kwargs)
while self.runsignal.is_set(): while self.runsignal.is_set():
@ -195,7 +196,8 @@ class Async_Optimize(object):
# print "^C" # print "^C"
self.runsignal.clear() self.runsignal.clear()
p.join() p.join()
if c.is_alive(): c.join()
if c and c.is_alive():
print "WARNING: callback still running, optimisation done!" print "WARNING: callback still running, optimisation done!"
return p.result return p.result
@ -208,11 +210,11 @@ class CGD(Async_Optimize):
if df returns tuple (grad, natgrad) it will optimize according if df returns tuple (grad, natgrad) it will optimize according
to natural gradient rules to natural gradient rules
''' '''
name = "Conjugate Gradient Descent" opt_name = "Conjugate Gradient Descent"
def fmin_async(self, *a, **kw): def opt_async(self, *a, **kw):
""" """
fmin_async(self, f, df, x0, callback, update_rule=FletcherReeves, opt_async(self, f, df, x0, callback, update_rule=FletcherReeves,
messages=0, maxiter=5e3, max_f_eval=15e3, gtol=1e-6, messages=0, maxiter=5e3, max_f_eval=15e3, gtol=1e-6,
report_every=10, *args, **kwargs) report_every=10, *args, **kwargs)
@ -240,11 +242,11 @@ class CGD(Async_Optimize):
at end of optimization! at end of optimization!
""" """
return super(CGD, self).fmin_async(*a, **kw) return super(CGD, self).opt_async(*a, **kw)
def fmin(self, *a, **kw): def opt(self, *a, **kw):
""" """
fmin(self, f, df, x0, callback=None, update_rule=FletcherReeves, opt(self, f, df, x0, callback=None, update_rule=FletcherReeves,
messages=0, maxiter=5e3, max_f_eval=15e3, gtol=1e-6, messages=0, maxiter=5e3, max_f_eval=15e3, gtol=1e-6,
report_every=10, *args, **kwargs) report_every=10, *args, **kwargs)
@ -267,5 +269,5 @@ class CGD(Async_Optimize):
at end of optimization at end of optimization
""" """
return super(CGD, self).fmin(*a, **kw) return super(CGD, self).opt(*a, **kw)

16
GPy/models/Bayesian_GPLVM.py
View file

@ -259,28 +259,28 @@ class Bayesian_GPLVM(sparse_GP, GPLVM):
ax2.text(.5, .5, r"$\mathbf{X}$", alpha=.5, transform=ax2.transAxes, ax2.text(.5, .5, r"$\mathbf{X}$", alpha=.5, transform=ax2.transAxes,
ha='center', va='center') ha='center', va='center')
figs[-1].canvas.draw() figs[-1].canvas.draw()
figs[-1].tight_layout(rect=(0, 0, 1, .9)) figs[-1].tight_layout(rect=(0, 0, 1, .86))
# ax3 = pylab.subplot2grid(splotshape, (3, 0), 2, 4, sharex=ax2) # ax3 = pylab.subplot2grid(splotshape, (3, 0), 2, 4, sharex=ax2)
figs.append(pylab.figure("BGPLVM DEBUG S", figsize=(12, 4))) figs.append(pylab.figure("BGPLVM DEBUG S", figsize=(12, 4)))
ax3 = self._debug_get_axis(figs) ax3 = self._debug_get_axis(figs)
ax3.text(.5, .5, r"$\mathbf{S}$", alpha=.5, transform=ax3.transAxes, ax3.text(.5, .5, r"$\mathbf{S}$", alpha=.5, transform=ax3.transAxes,
ha='center', va='center') ha='center', va='center')
figs[-1].canvas.draw() figs[-1].canvas.draw()
figs[-1].tight_layout(rect=(0, 0, 1, .9)) figs[-1].tight_layout(rect=(0, 0, 1, .86))
# ax4 = pylab.subplot2grid(splotshape, (5, 0), 2, 2) # ax4 = pylab.subplot2grid(splotshape, (5, 0), 2, 2)
figs.append(pylab.figure("BGPLVM DEBUG Z", figsize=(6, 4))) figs.append(pylab.figure("BGPLVM DEBUG Z", figsize=(6, 4)))
ax4 = self._debug_get_axis(figs) ax4 = self._debug_get_axis(figs)
ax4.text(.5, .5, r"$\mathbf{Z}$", alpha=.5, transform=ax4.transAxes, ax4.text(.5, .5, r"$\mathbf{Z}$", alpha=.5, transform=ax4.transAxes,
ha='center', va='center') ha='center', va='center')
figs[-1].canvas.draw() figs[-1].canvas.draw()
figs[-1].tight_layout(rect=(0, 0, 1, .9)) figs[-1].tight_layout(rect=(0, 0, 1, .86))
# ax5 = pylab.subplot2grid(splotshape, (5, 2), 2, 2) # ax5 = pylab.subplot2grid(splotshape, (5, 2), 2, 2)
figs.append(pylab.figure("BGPLVM DEBUG theta", figsize=(6, 4))) figs.append(pylab.figure("BGPLVM DEBUG theta", figsize=(6, 4)))
ax5 = self._debug_get_axis(figs) ax5 = self._debug_get_axis(figs)
ax5.text(.5, .5, r"${\theta}$", alpha=.5, transform=ax5.transAxes, ax5.text(.5, .5, r"${\theta}$", alpha=.5, transform=ax5.transAxes,
ha='center', va='center') ha='center', va='center')
figs[-1].canvas.draw() figs[-1].canvas.draw()
figs[-1].tight_layout(rect=(0, 0, 1, .9)) figs[-1].tight_layout(rect=(.15, 0, 1, .86))
figs.append(pylab.figure("BGPLVM DEBUG Kmm", figsize=(12, 6))) figs.append(pylab.figure("BGPLVM DEBUG Kmm", figsize=(12, 6)))
fig = figs[-1] fig = figs[-1]
ax6 = fig.add_subplot(121) ax6 = fig.add_subplot(121)
@ -345,16 +345,16 @@ class Bayesian_GPLVM(sparse_GP, GPLVM):
# loc=3, ncol=self.Q, bbox_to_anchor=(0, 1.15, 1, 1.15), # loc=3, ncol=self.Q, bbox_to_anchor=(0, 1.15, 1, 1.15),
# borderaxespad=0, mode="expand") # borderaxespad=0, mode="expand")
ax2.legend(Xlatentplts, [r"$Q_{}$".format(i + 1) for i in range(self.Q)], ax2.legend(Xlatentplts, [r"$Q_{}$".format(i + 1) for i in range(self.Q)],
loc=3, ncol=self.Q, bbox_to_anchor=(0, 1.01, 1, 1.01), loc=3, ncol=self.Q, bbox_to_anchor=(0, 1.1, 1, 1.1),
borderaxespad=0, mode="expand") borderaxespad=0, mode="expand")
ax3.legend(Xlatentplts, [r"$Q_{}$".format(i + 1) for i in range(self.Q)], ax3.legend(Xlatentplts, [r"$Q_{}$".format(i + 1) for i in range(self.Q)],
loc=3, ncol=self.Q, bbox_to_anchor=(0, 1.01, 1, 1.01), loc=3, ncol=self.Q, bbox_to_anchor=(0, 1.1, 1, 1.1),
borderaxespad=0, mode="expand") borderaxespad=0, mode="expand")
ax4.legend(Xlatentplts, [r"$Q_{}$".format(i + 1) for i in range(self.Q)], ax4.legend(Xlatentplts, [r"$Q_{}$".format(i + 1) for i in range(self.Q)],
loc=3, ncol=self.Q, bbox_to_anchor=(0, 1.01, 1, 1.01), loc=3, ncol=self.Q, bbox_to_anchor=(0, 1.1, 1, 1.1),
borderaxespad=0, mode="expand") borderaxespad=0, mode="expand")
ax5.legend(Xlatentplts, [r"$Q_{}$".format(i + 1) for i in range(self.Q)], ax5.legend(Xlatentplts, [r"$Q_{}$".format(i + 1) for i in range(self.Q)],
loc=3, ncol=self.Q, bbox_to_anchor=(0, 1.01, 1, 1.01), loc=3, ncol=self.Q, bbox_to_anchor=(0, 1.1, 1, 1.1),
borderaxespad=0, mode="expand") borderaxespad=0, mode="expand")
Lleg = ax1.legend() Lleg = ax1.legend()
Lleg.draggable() Lleg.draggable()

6
GPy/testing/cgd_tests.py
View file

@ -26,7 +26,7 @@ class Test(unittest.TestCase):
for _ in range(restarts): for _ in range(restarts):
try: try:
x0 = numpy.random.randn(N) * .5 x0 = numpy.random.randn(N) * .5
res = opt.fmin(f, df, x0, messages=0, res = opt.opt(f, df, x0, messages=0,
maxiter=1000, gtol=1e-10) maxiter=1000, gtol=1e-10)
assert numpy.allclose(res[0], 0, atol=1e-3) assert numpy.allclose(res[0], 0, atol=1e-3)
break break
@ -48,7 +48,7 @@ class Test(unittest.TestCase):
for _ in range(restarts): for _ in range(restarts):
try: try:
x0 = numpy.random.randn(N) * .5 x0 = numpy.random.randn(N) * .5
res = opt.fmin(f, df, x0, messages=0, res = opt.opt(f, df, x0, messages=0,
maxiter=1000, gtol=1e-2) maxiter=1000, gtol=1e-2)
assert numpy.allclose(res[0], 1, atol=.01) assert numpy.allclose(res[0], 1, atol=.01)
break break
@ -103,7 +103,7 @@ if __name__ == "__main__":
if r[-1] != RUNNING: if r[-1] != RUNNING:
res[0] = r res[0] = r
p, c = opt.fmin_async(f, df, x0.copy(), callback, messages=True, maxiter=1000, p, c = opt.opt_async(f, df, x0.copy(), callback, messages=True, maxiter=1000,
report_every=20, gtol=1e-12) report_every=20, gtol=1e-12)