From 79748e3c13390e39306be7373c0536c43bf8b117 Mon Sep 17 00:00:00 2001
From: "Alinson S. Xavier" <git@axavier.org>
Date: Tue, 24 Jan 2023 10:27:41 -0600
Subject: [PATCH] Dist: Drop NaN in training dataset

---
 src/graph/dist.jl | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/src/graph/dist.jl b/src/graph/dist.jl
index 777ad87..a5e1abf 100644
--- a/src/graph/dist.jl
+++ b/src/graph/dist.jl
@@ -40,7 +40,8 @@ function _calculate_distance(
         end
 
         # Fit kNN model
-        df = DataFrame(CSV.File(csv_filename))
+        df = DataFrame(CSV.File(csv_filename, missingstring="NaN"))
+        dropmissing!(df)
         coords = Matrix(df[!, [:source_lat, :source_lon, :dest_lat, :dest_lon]])'
         metric.ratios = Matrix(df[!, [:ratio]])
         metric.tree = KDTree(coords)
@@ -53,5 +54,7 @@ function _calculate_distance(
     # Predict ratio
     idxs, _ = knn(metric.tree, [source_lat, source_lon, dest_lat, dest_lon], 5)
     ratio_pred = mean(metric.ratios[idxs])
-    return round(dist_euclidean * ratio_pred, digits = 3)
+    dist_pred = round(dist_euclidean * ratio_pred, digits = 3)
+    isfinite(dist_pred) || error("non-finite distance detected: $dist_pred")
+    return dist_pred
 end