apunkt/GPy
1
0
Fork 0
mirror of https://github.com/SheffieldML/GPy.git synced 2026-05-10 20:42:39 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

update string checks in initialization.py

Browse source
This commit is contained in:
Martin Bubel 2024-07-21 16:14:07 +02:00
parent bfa1d67fda
commit 92fb49d001
1 changed files with 15 additions and 9 deletions
Download patch file Download diff file Expand all files Collapse all files

24
GPy/util/initialization.py
View file

@ -1,26 +1,32 @@
''' """
Created on 24 Feb 2014 Created on 24 Feb 2014
@author: maxz @author: maxz
''' """
import numpy as np import numpy as np
from ..util.pca import PCA from ..util.pca import PCA
def initialize_latent(init, input_dim, Y): def initialize_latent(init, input_dim, Y):
Xr = np.asfortranarray(np.random.normal(0, 1, (Y.shape[0], input_dim))) Xr = np.asfortranarray(np.random.normal(0, 1, (Y.shape[0], input_dim)))
if 'PCA' in init: if "PCA" == init:
p = PCA(Y) p = PCA(Y)
PC = p.project(Y, min(input_dim, Y.shape[1])) PC = p.project(Y, min(input_dim, Y.shape[1]))
Xr[:PC.shape[0], :PC.shape[1]] = PC Xr[: PC.shape[0], : PC.shape[1]] = PC
var = .1*p.fracs[:input_dim] var = 0.1 * p.fracs[:input_dim]
elif init in 'empirical_samples': elif init == "empirical_samples":
from ..util.linalg import tdot from ..util.linalg import tdot
from ..util import diag from ..util import diag
YYT = tdot(Y) YYT = tdot(Y)
diag.add(YYT, 1e-6) diag.add(YYT, 1e-6)
EMP = np.asfortranarray(np.random.multivariate_normal(np.zeros(Y.shape[0]), YYT, min(input_dim, Y.shape[1])).T) EMP = np.asfortranarray(
Xr[:EMP.shape[0], :EMP.shape[1]] = EMP np.random.multivariate_normal(
np.zeros(Y.shape[0]), YYT, min(input_dim, Y.shape[1])
).T
)
Xr[: EMP.shape[0], : EMP.shape[1]] = EMP
var = np.random.uniform(0.5, 1.5, input_dim) var = np.random.uniform(0.5, 1.5, input_dim)
else: else:
var = Xr.var(0) var = Xr.var(0)
@ -28,4 +34,4 @@ def initialize_latent(init, input_dim, Y):
Xr -= Xr.mean(0) Xr -= Xr.mean(0)
Xr /= Xr.std(0) Xr /= Xr.std(0)
return Xr, var/var.max() return Xr, var / var.max()