Format workspace

src/fastcpd.cc

@@ -119,6 +119,28 @@ arma::colvec cost_update_gradient(
   return gradient;
 }
 
+arma::mat cost_update_hessian(
+    arma::mat data,
+    arma::colvec theta,
+    std::string family,
+    double min_prob
+) {
+  arma::rowvec new_data = data.row(data.n_rows - 1);
+  arma::rowvec x = new_data.tail(new_data.n_elem - 1);
+  arma::mat hessian;
+  if (family.compare("binomial") == 0) {
+    double prob = 1 / (1 + exp(-arma::as_scalar(x * theta)));
+    hessian = (x.t() * x) * arma::as_scalar((1 - prob) * prob);
+  } else if (family.compare("poisson") == 0) {
+    double prob = exp(arma::as_scalar(x * theta));
+    hessian = (x.t() * x) * std::min(arma::as_scalar(prob), min_prob);
+  } else {
+    // `family` is either "lasso" or "gaussian".
+    hessian = x.t() * x;
+  }
+  return hessian;
+}
+
 Rcpp::List cost_update(
     const arma::mat data,
     arma::mat theta_hat,

src/cost_update_hessian.cc → src/fastcpd_test.cc

@@ -1,27 +1,5 @@
 #include "fastcpd.h"
 
-arma::mat cost_update_hessian(
-    arma::mat data,
-    arma::colvec theta,
-    std::string family,
-    double min_prob
-) {
-  arma::rowvec new_data = data.row(data.n_rows - 1);
-  arma::rowvec x = new_data.tail(new_data.n_elem - 1);
-  arma::mat hessian;
-  if (family.compare("binomial") == 0) {
-    double prob = 1 / (1 + exp(-arma::as_scalar(x * theta)));
-    hessian = (x.t() * x) * arma::as_scalar((1 - prob) * prob);
-  } else if (family.compare("poisson") == 0) {
-    double prob = exp(arma::as_scalar(x * theta));
-    hessian = (x.t() * x) * std::min(arma::as_scalar(prob), min_prob);
-  } else {
-    // `family` is either "lasso" or "gaussian".
-    hessian = x.t() * x;
-  }
-  return hessian;
-}
-
 context("cost_update_hessian Unit Test") {
   test_that("binomal is correct for a two dimensional data") {
     arma::colvec theta = {-0.5, 0.3};