Github user myui commented on a diff in the pull request:
https://github.com/apache/incubator-hivemall/pull/149#discussion_r205390805
--- Diff: core/src/main/java/hivemall/fm/FactorizationMachineModel.java ---
@@ -399,9 +399,8 @@ public void initRandom(int factor, long seed) {
protected static final void uniformFill(final float[] a, final Random
rand,
final float maxInitValue) {
final int len = a.length;
- final float basev = maxInitValue / len;
for (int i = 0; i < len; i++) {
- float v = rand.nextFloat() * basev;
+ float v = rand.nextFloat() * maxInitValue;
--- End diff --
While this modified `random` initialization is not used for classification
(and only for regression), your evaluation is only for classification.
This, it's doubtful that this change contributed for improving accuracy.
---
