Transforming variables in multi-logit regression

I am trying to fit a multi-logit regression using categorical_logit(). I have read the example in the documentation ( but instead of reading in my regressors as a matrix I have input them individually because I want to transform them. However, I get the error “No matches for available argument signatures for categorical_logit: Real return type required for probability function.” I am not sure how to alter the format of my input to categorical_logit () while maintaining the variable transformation. Any advice is appreciated.

  int ntr;          // N
  int nsub;         
  int ParticipantID[ntr];
  int nResponseTypes; // K
  real x1[ntr];
  real x2[ntr];
  real x3[ntr];
  //real x4[ntr]; //uses same input data as x3
  int actionTaken[ntr];  

  real<lower=0> invTemp[nsub]; 
  real b_x1_pr[nsub];
  real b_x2_pr[nsub];
  real b_x3_pr[nsub];

transformed parameters {
  real k[nsub];
  real b_x1[nsub];
  real b_x2[nsub];
  real b_x3[nsub];
  real b_x4[nsub];

  for (is in 1:nsub){
    k[is] = fabs(1) + fabs(b_x1_pr[is])+fabs(b_x2_pr[is])+fabs(b_x3_pr[is]);
    b_x4[is]               = 1/k[is];
    b_x1[is]               = b_x1_pr[is]/k[is];
    b_x2[is]               = b_x2_pr[is]/k[is];
    b_x3[is]               = b_x3_pr[is]/k[is];

  matrix[ntr,nResponseTypes] invTxUtil;
  // Priors
  invTemp ~ cauchy(10,2);
  b_x1_pr ~ cauchy(0,1);
  b_x2_pr ~ cauchy(0,1);
  b_x3_pr ~ cauchy(0,1);

  for (itr in 1:ntr) {
    // Utility 1
    invTxUtil[itr,1] = invTemp[ParticipantID[itr]]*(b_x1[ParticipantID[itr]]*x1[itr]);
    // Utility 2
    invTxUtil[itr,2] = invTemp[ParticipantID[itr]]*(b_x4[ParticipantID[itr]]*x3[itr]);
    // Utility 3
    invTxUtil[itr,3] = invTemp[ParticipantID[itr]]*(b_x2[ParticipantID[itr]]*x2[itr] + b_x3[ParticipantID[itr]]*x3[itr]);

  actionTaken ~ categorical_logit(invTxUtil);

Looks like categorical_logit is a bit picky about what types it accepts. This works.

for (i in 1:ntr)
    actionTaken[i] ~ categorical_logit(to_vector(invTxUtil[i]));