问题
I have a convex function f(b):
f(b) = log(pi) - lambda * log( t(r) %*% b)
dim(pi) = (n,1), lambda is a scalar constant, dim(r) = (n,n)
b is the parameter, dim(b) = (n,1)
The call to
is_convex(f)
is_dcp(f)
both return TRUE.
However, I don’t understand why the following returns FALSE:
is_dcp(log_sum_exp(f))
As far as I understand, log_sum_exp of a convex function is convex?
I noticed that both
is_convex(f)
and
is_concave(f)
return FALSE.
So it seems like CVRX is unable to deduce the convexity of log_sum_exp(f)?
Any help would be appreciated
(I use CVRX ver 0.99 package in R3.5)
回答1:
This bug has been fixed in CVXR v0.99-5.
来源:https://stackoverflow.com/questions/50615864/log-sum-exp-of-a-convex-function-is-not-dcp-compliant