move data generation into the epoch loop

Previously we would spend all of the training epochs on making the network a better predictor of the *initial* TD-leaf values that we computed before the start of the first epoch. There's definitely something stale about this, given that the TD-leaf training targets involved the net evaluation at PVs. I'm not overly confident that this makes sense, but as a kind of compromise between fully on-line TD-leaf and the fully stale version we had before, we now recompute the TD-leaf values in each epoch with the fine-tuned net instead of the epoch 0 net.

Perhaps we were just lucky, but this seems to train a strong network under the default setting.

More experimentation is required.

1 files changed, 1 insertions, 1 deletions

diff --git a/Makefile b/Makefile
index 1b87677..d6ca030 100644
--- a/Makefile
+++ b/Makefile
@@ -37,7 +37,7 @@ cand_b = $(word 2,$(subst -, ,$1))
 .PHONY: selfplay-%
 selfplay-%:
 	rm -f selfplay-%
-	rm -f src/*.o include/*.o
+	rm -f src/*.o include/negamax.o include/pv.o include/tt.o
 	$(MAKE) SELFPLAY_OUT=$@ \
 	  DEFINES="-DRECORD_PV -DCAND_A=$(call cand_a,$*) -DCAND_B=$(call cand_b,$*) \
 				-DCAND_A_WEIGHTS_5='\"cand_$(call cand_a,$*)_weights_5.c\"' \