[ML] Fixing categorization tokens for multi-line messages (#103007)

jgowdyelastic · kibanamachine · web-flow · commit 824463ace594 · 2021-06-29T10:28:51.000+01:00
Co-authored-by: Kibana Machine &lt;42973632+kibanamachine@users.noreply.github.com&gt;
diff --git a/x-pack/plugins/ml/server/models/job_service/new_job/categorization/examples.ts b/x-pack/plugins/ml/server/models/job_service/new_job/categorization/examples.ts
@@ -145,10 +145,11 @@ export function categorizationExamplesProvider({
         for (let g = 0; g < sumLengths.length; g++) {
           if (t.start_offset <= sumLengths[g] + g) {
             const offset = g > 0 ? sumLengths[g - 1] + g : 0;
+            const start = t.start_offset - offset;
             tokensPerExample[g].push({
               ...t,
-              start_offset: t.start_offset - offset,
-              end_offset: t.end_offset - offset,
+              start_offset: start,
+              end_offset: start + t.token.length,
             });
             break;
           }
