apunkt/GPy
1
0
Fork 0
mirror of https://github.com/SheffieldML/GPy.git synced 2026-05-05 17:52:39 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

[var dtc missing] no caching of indexed y anymore, no subarray indexing

Browse source
This commit is contained in:
mzwiessele 2014-07-02 11:04:59 -07:00
parent a9443417d7
commit 4a8a9bac54
3 changed files with 19 additions and 31 deletions
Download patch file Download diff file Expand all files Collapse all files

2
GPy/examples/dimensionality_reduction.py
View file

@ -296,7 +296,7 @@ def bgplvm_simulation_missing_data(optimize=True, verbose=1,
from GPy.models import BayesianGPLVM from GPy.models import BayesianGPLVM
from GPy.inference.latent_function_inference.var_dtc import VarDTCMissingData from GPy.inference.latent_function_inference.var_dtc import VarDTCMissingData
D1, D2, D3, N, num_inducing, Q = 6, 5, 8, 400, 3, 4 D1, D2, D3, N, num_inducing, Q = 1000, 5, 8, 400, 3, 4
_, _, Ylist = _simulate_sincos(D1, D2, D3, N, num_inducing, Q, plot_sim) _, _, Ylist = _simulate_sincos(D1, D2, D3, N, num_inducing, Q, plot_sim)
Y = Ylist[0] Y = Ylist[0]
k = kern.Linear(Q, ARD=True)# + kern.white(Q, _np.exp(-2)) # + kern.bias(Q) k = kern.Linear(Q, ARD=True)# + kern.white(Q, _np.exp(-2)) # + kern.bias(Q)

46
GPy/inference/latent_function_inference/var_dtc.py
View file

@ -219,40 +219,26 @@ class VarDTCMissingData(LatentFunctionInference):
if self._inan is None: if self._inan is None:
inan = np.isnan(Y) inan = np.isnan(Y)
has_none = inan.any() has_none = inan.any()
self._inan = inan
else: else:
inan = self._inan inan = self._inan
has_none = True has_none = True
if has_none: if has_none:
print "caching missing data slices, this can take several minutes depending on the number of unique dimensions of the data..." #print "caching missing data slices, this can take several minutes depending on the number of unique dimensions of the data..."
from ...util.subarray_and_sorting import common_subarrays #csa = common_subarrays(inan, 1)
self._subarray_indices = [] size = Y.shape[1]
csa = common_subarrays(inan, 1)
size = len(csa)
next_ten = 0
for i, (v,ind) in enumerate(csa.iteritems()):
if not np.all(v):
if ((i+1.)/size) >= next_ten:
logger.info('preparing subarrays {:3%}'.format((i+1.)/size))
next_ten += max(.1, 1./size)
v = ~np.array(v, dtype=bool)
ind = np.array(ind, dtype=int)
if ind.size == Y.shape[1]:
ind = slice(None)
self._subarray_indices.append([v,ind])
#logger.info('preparing subarrays {:3.3%}'.format((i+1.)/size)) #logger.info('preparing subarrays {:3.3%}'.format((i+1.)/size))
#Ys = [Y[v, :][:, ind] for v, ind in self._subarray_indices] #Ys = [Y[v, :][:, ind] for v, ind in self._subarray_indices]
logger.info('preparing traces Y')
next_ten = [0.] next_ten = [0.]
count = itertools.count() count = itertools.count()
def trace(y, v, ind): def trace(y):
i = count.next() i = count.next()
if ((i+1.)/size) >= next_ten[0]: if ((i+1.)/size) >= next_ten[0]:
logger.info('preparing traces {:3%}'.format((i+1.)/size)) logger.info('preparing traces {:> 3.1%}'.format((i+1.)/size))
next_ten[0] += .1 next_ten[0] += .1
y = y[v,:][:,ind] y = y[~inan[:,i],i:i+1]
return np.einsum('ij,ij->', y,y) return np.einsum('ij,ij->', y,y)
traces = [trace(Y, v, ind) for v, ind in self._subarray_indices] traces = [trace(Y) for _ in xrange(size)]
return traces return traces
else: else:
self._subarray_indices = [[slice(None),slice(None)]] self._subarray_indices = [[slice(None),slice(None)]]
@ -301,14 +287,16 @@ class VarDTCMissingData(LatentFunctionInference):
#logger.info('computing dimension-wise likelihood and derivatives') #logger.info('computing dimension-wise likelihood and derivatives')
#size = len(Ys) #size = len(Ys)
for trYYT, [v, ind] in itertools.izip(traces, self._subarray_indices): for i, trYYT in enumerate(traces):
y = Y[v,:][:,ind] v = ~self._inan[:, i]
y = Y[v,i:i+1]#[:,ind]
#y = Y[v]
#logger.info('{:.3%} dimensions:{}'.format((i+1.)/size, ind)) #logger.info('{:.3%} dimensions:{}'.format((i+1.)/size, ind))
if het_noise: beta = beta_all[ind] if het_noise: beta = beta_all[i]
else: beta = beta_all else: beta = beta_all
VVT_factor = (beta*y) VVT_factor = (y*beta)
output_dim = y.shape[1] output_dim = 1#len(ind)
psi0 = psi0_all[v] psi0 = psi0_all[v]
psi1 = psi1_all[v, :] psi1 = psi1_all[v, :]
@ -368,7 +356,7 @@ class VarDTCMissingData(LatentFunctionInference):
data_fit, num_data, output_dim, trYYT, Y) data_fit, num_data, output_dim, trYYT, Y)
#if full_VVT_factor: #if full_VVT_factor:
woodbury_vector[:, ind] = Cpsi1Vf woodbury_vector[:, i:i+1] = Cpsi1Vf
#else: #else:
# print 'foobar' # print 'foobar'
# tmp, _ = dtrtrs(Lm, psi1V, lower=1, trans=0) # tmp, _ = dtrtrs(Lm, psi1V, lower=1, trans=0)
@ -380,7 +368,7 @@ class VarDTCMissingData(LatentFunctionInference):
symmetrify(Bi) symmetrify(Bi)
Bi = -dpotri(LB, lower=1)[0] Bi = -dpotri(LB, lower=1)[0]
diag.add(Bi, 1) diag.add(Bi, 1)
woodbury_inv_all[:, :, ind] = backsub_both_sides(Lm, Bi)[:,:,None] woodbury_inv_all[:, :, i:i+1] = backsub_both_sides(Lm, Bi)[:,:,None]
dL_dthetaL = likelihood.exact_inference_gradients(dL_dR) dL_dthetaL = likelihood.exact_inference_gradients(dL_dR)

2
GPy/models/bayesian_gplvm.py
View file

@ -37,7 +37,7 @@ class BayesianGPLVM(SparseGP):
self.init = init self.init = init
if X_variance is None: if X_variance is None:
self.logger.info("initializing latent space variance ~ N(0,.1)") self.logger.info("initializing latent space variance ~ uniform(0,.1)")
X_variance = np.random.uniform(0,.1,X.shape) X_variance = np.random.uniform(0,.1,X.shape)
if Z is None: if Z is None: