cleanup, match now plays pairs of symmetric games

1 files changed, 6 insertions, 21 deletions

diff --git a/src/nn_train.py b/src/nn_train.py
index cde7a98..8f1a73e 100755
--- a/src/nn_train.py
+++ b/src/nn_train.py
@@ -201,8 +201,7 @@ class SelfPlayRecords:
 class Net(Module):
     def __init__(self, inp_dim: int, hidden: int) -> None:
         super().__init__()
-        # learned 7-way cell-state embedding (empty, {black, white} x
-        # {flat, standing, road}); the slot layout must match nn.c
+        # learned cell-state embedding, see nn.c
         self.embedding = Embedding(7, 1)
         self.hidden = Linear(inp_dim, hidden, bias=True)
         self.output = Linear(hidden, 1, bias=True)
@@ -302,7 +301,7 @@ class Net(Module):
                 _ = f.write(f"/* {p} */\n")
             _ = f.write(f'#include "weights_{size}.h"\n\n')
             _ = f.write(
-                f"const float {prefix}_embedding[7] =\n"
+                f"const float {prefix}_embedding[{len(E)}] =\n"
                 f"{{\n{_fmt_flat(E, per_line=7)}\n}};\n\n"
             )
             _ = f.write(
@@ -548,24 +547,6 @@ def main() -> None:
         help="number of previous self-play instances to mix in with current dataset during training",
     )
     _ = p.add_argument(
-        "--eps-start",
-        type=float,
-        default=0.15,
-        help="initial self-play epsilon: in asymmetric self-play one side (chosen at random per game) plays a uniform random move with this probability while the greedy side has its plies recorded",
-    )
-    _ = p.add_argument(
-        "--eps-min",
-        type=float,
-        default=0.01,
-        help="self-play epsilon floor",
-    )
-    _ = p.add_argument(
-        "--eps-decay",
-        type=float,
-        default=0.9,
-        help="per-iteration epsilon decay",
-    )
-    _ = p.add_argument(
         "--td-lambda",
         type=float,
         default=0.7,
@@ -639,6 +620,10 @@ def main() -> None:
 
     (root / "data").mkdir(exist_ok=True)
     champ = setup(size=args.size, root=root)
+
+    num_params = sum(p.numel() for p in champ.parameters() if p.requires_grad)
+    print(f"Number of trainable parameters: {num_params:,}")
+
     for iter in range(1, args.iters + 1):
         champ = run_step(
             batch=args.batch,