Github user myui commented on a diff in the pull request:
https://github.com/apache/incubator-hivemall/pull/149#discussion_r200590772
--- Diff: core/src/main/java/hivemall/fm/FactorizationMachineUDTF.java ---
@@ -283,9 +293,16 @@ public void process(Object[] args) throws
HiveException {
}
++_t;
- recordTrain(x, y);
- boolean adaptiveRegularization = (_va_rand != null) && _t >=
_validationThreshold;
- train(x, y, adaptiveRegularization);
+
+ boolean validation = false;
+ if ((_va_rand != null) && _t >= _validationThreshold) {
+ final float rnd = _va_rand.nextFloat();
+ validation = rnd < _validationRatio;
+ }
+
+ recordTrain(x, y, validation);
+
+ train(x, y, validation);
--- End diff --
Validation examples are fixed in this implementation. Also, not using
non-validation examples for regularization is a bad strategy.
---
