apunkt/GPy
1
0
Fork 0
mirror of https://github.com/SheffieldML/GPy.git synced 2026-05-09 20:12:38 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

Modifications to transformations ... not sure which tests to run to make sure I haven't messed things up. New code avoids exponentiating values greater than -log(eps) or less than log(eps). Also changed negative code to call the positive code (I think they should inherit the positive code ... but maybe not.

Browse source
This commit is contained in:
Neil Lawrence 2013-08-16 18:28:16 +01:00
parent 9e0795afc4
commit 7aac769d37
5 changed files with 58 additions and 28 deletions
Download patch file Download diff file Expand all files Collapse all files

41
GPy/core/transformations.py
View file

@ -4,6 +4,8 @@
import numpy as np
from GPy.core.domains import POSITIVE, NEGATIVE, BOUNDED
import sys
lim_val = -np.log(sys.float_info.epsilon)
class transformation(object):
domain = None
@ -17,7 +19,7 @@ class transformation(object):
""" df_dx evaluated at self.f(x)=f"""
raise NotImplementedError
def initialize(self, f):
""" produce a sensible initial values for f(x)"""
""" produce a sensible initial value for f(x)"""
raise NotImplementedError
def __str__(self):
raise NotImplementedError
@ -25,13 +27,14 @@ class transformation(object):
class logexp(transformation):
domain = POSITIVE
def f(self, x):
return np.log(1. + np.exp(x))
return np.where(x>lim_val, x, np.log(1. + np.exp(x)))
def finv(self, f):
return np.log(np.exp(f) - 1.)
return np.where(f>lim_val, f, np.log(np.exp(f) - 1.))
def gradfactor(self, f):
ef = np.exp(f)
return (ef - 1.) / ef
return np.where(f>lim_val, 1., 1 - np.exp(-f))
def initialize(self, f):
if np.any(f < 0.):
print "Warning: changing parameters to satisfy constraints"
return np.abs(f)
def __str__(self):
return '(+ve)'
@ -39,18 +42,19 @@ class logexp(transformation):
class negative_logexp(transformation):
domain = NEGATIVE
def f(self, x):
return -np.log(1. + np.exp(x))
return -logexp.f(x) #np.log(1. + np.exp(x))
def finv(self, f):
return np.log(np.exp(-f) - 1.)
return logexp.finv(-f) #np.log(np.exp(-f) - 1.)
def gradfactor(self, f):
ef = np.exp(-f)
return -(ef - 1.) / ef
return -logexp.gradfactor(-f)
#ef = np.exp(-f)
#return -(ef - 1.) / ef
def initialize(self, f):
return -np.abs(f)
return -logexp.initialize(f) #np.abs(f)
def __str__(self):
return '(-ve)'
class logexp_clipped(transformation):
class logexp_clipped(logexp):
max_bound = 1e100
min_bound = 1e-10
log_max_bound = np.log(max_bound)
@ -78,9 +82,10 @@ class logexp_clipped(transformation):
return '(+ve_c)'
class exponent(transformation):
# TODO: can't allow this to go to zero, need to set a lower bound. Similar with negative exponent below. See old MATLAB code.
domain = POSITIVE
def f(self, x):
return np.exp(x)
return np.where(x<lim_val, np.where(x>-lim_val, np.exp(x), np.exp(-lim_val)), np.exp(lim_val))
def finv(self, x):
return np.log(x)
def gradfactor(self, f):
@ -92,18 +97,16 @@ class exponent(transformation):
def __str__(self):
return '(+ve)'
class negative_exponent(transformation):
class negative_exponent(exponent):
domain = NEGATIVE
def f(self, x):
return -np.exp(x)
def finv(self, x):
return np.log(-x)
return -exponent.f(x)
def finv(self, f):
return exponent.finv(-f)
def gradfactor(self, f):
return f
def initialize(self, f):
if np.any(f > 0.):
print "Warning: changing parameters to satisfy constraints"
return -np.abs(f)
return -exponent.initialize(f) #np.abs(f)
def __str__(self):
return '(-ve)'