From d238d2954e83f6c18ea87bd6b5aee78254e79770 Mon Sep 17 00:00:00 2001 From: Jens Ahrensfeld Date: Mon, 5 Jan 2026 19:40:05 +0100 Subject: [PATCH] grad_compute(): pre-multiply weight decay with learning rate --- src/rbm/entity.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/src/rbm/entity.py b/src/rbm/entity.py index bf3c5e2..68c6f74 100644 --- a/src/rbm/entity.py +++ b/src/rbm/entity.py @@ -98,7 +98,7 @@ class Entity: # Compute gradient self.grad.b_v = (momentum * self.grad.b_v + learning_rate * d_bv) self.grad.b_h = (momentum * self.grad.b_h + learning_rate * d_bh) - self.grad.w_hv = (momentum * self.grad.w_hv + learning_rate * d_whv - weight_decay * self.state.w_hv) + self.grad.w_hv = (momentum * self.grad.w_hv + learning_rate * d_whv - learning_rate * weight_decay * self.state.w_hv) return self.grad