From 8aa274d2964a199f9cf81fe6e5d205c03549ff9c Mon Sep 17 00:00:00 2001
From: "Brian C. Van Essen" <vanessen1@llnl.gov>
Date: Tue, 25 Jun 2024 11:43:36 -0700
Subject: [PATCH] Increasing the precision of the reported error for check
 metric.

---
 src/callbacks/check_metric.cpp | 15 ++++++++++++---
 1 file changed, 12 insertions(+), 3 deletions(-)

diff --git a/src/callbacks/check_metric.cpp b/src/callbacks/check_metric.cpp
index 611e2b16bfc..b3e8f8f79cd 100644
--- a/src/callbacks/check_metric.cpp
+++ b/src/callbacks/check_metric.cpp
@@ -37,6 +37,7 @@
 
 #include "lbann/proto/callbacks.pb.h"
 
+#include <iomanip>
 #include <set>
 #include <sstream>
 #include <string>
@@ -59,11 +60,19 @@ check_metric::check_metric(std::string metric_name,
   if (lower_bound > upper_bound) {
     std::stringstream err;
     err << "callback \"" << name() << "\" "
-        << "got an invalid range for metric values "
+        << "got an invalid range for metric values " << std::setprecision(9)
         << "(lower bound " << m_lower_bound << ", "
         << "upper bound " << m_upper_bound << ")";
     LBANN_ERROR(err.str());
   }
+  if (lower_bound == upper_bound) {
+    std::stringstream err;
+    err << "callback \"" << name() << "\" "
+        << "got an zero range for metric values " << std::setprecision(9)
+        << "(lower bound " << m_lower_bound << " == "
+        << "upper bound " << m_upper_bound << ")";
+    LBANN_WARNING(err.str());
+  }
 }
 
 check_metric::check_metric() : check_metric("", {}, 0, 0, false) {}
@@ -122,8 +131,8 @@ void check_metric::do_check_metric(const model& m) const
   if (!(m_lower_bound <= value && value <= m_upper_bound)) {
     err << "callback \"" << name() << "\" expected "
         << "metric \"" << m_metric_name << "\" "
-        << "to have a value in range "
-        << "[" << m_lower_bound << "," << m_upper_bound << "], "
+        << "to have a value in range " << std::setprecision(9) << "["
+        << m_lower_bound << "," << m_upper_bound << "], "
         << "but found a value of " << value;
     if (m_error_on_failure) {
       LBANN_ERROR(err.str());