Casadi MX type softMax math operation

[Guang Yang]

Hello,

I am trying to apply a softmax function on a MX type matrix with (n x 1) dimension.

from casadi import *
import math

n=10 #dimension
mx_vec = MX.sym('mx_vec',n,1)

def softMax():
    sum = 0
    C = 5.0 #C is a constant for softmax function
    for mx_element in mx_vec:
        sum = sum + math.exp(mx_element*C)
    max = math.log(sum)/C
    return max

But I got a runtime error

RuntimeError:.../casadi/core/mx_node.cpp:373:'to_double' not defined for class N6casadi7UnaryMXE

The softMax() function works for a list of float type when I test it. But how can I implement it with MX type?

[Joris Gillis]

Dear Guang,

No need to iterate over elements, and no need for 'math':

log(sum1(exp(mx_vec*C)))/C

Best,

 Joris