TD-leaf and another negamax bug

TD-lambda and TD-leaf are separated by the observation that minimax with a parameterised leaf heuristic is differentiable* in the parameters. "All" that had to be done was to extract the "principal variation": which one position survived all of the pruning to become the reported value.

This was a bit fiddly and uncovered a subtle bug (again) in negamax.c: we were not being complete in our overriding of the TT entry and by omitting key it was possible that later searches would find incorrect values.

The more interesting question to me was: how do we introduce enough variety while not corrupting the PV chain? I settled on having selfplay have one of the players be epsilon-random but the other greedy (minimax) and simply report only the greedy portion PV chain. This appears to work well enough in practice and doesn't introduce nonsense training data for the net (we don't force its actions to be nonsensical). I also deciced, perhaps in a departure from tradition, that we had good supervised labels for PVs that happened to be winning states, so the training overrides the targets in that case. Finally rolled back some of the ambition in nn_train.py default arguments.

*almost always

1 files changed, 0 insertions, 54 deletions

diff --git a/include/nn_input.c b/include/nn_input.c
deleted file mode 100644
index 2fbd8da..0000000
--- a/include/nn_input.c
+++ /dev/null
@@ -1,54 +0,0 @@
-/*
-  This file is part of ctak.
-
-  This program is free software: you can redistribute it and/or modify
-  it under the terms of the GNU General Public License as published by
-  the Free Software Foundation, either version 3 of the License, or
-  (at your option) any later version.
-
-  This program is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
-  General Public License for more details.
-
-  You should have received a copy of the GNU General Public License
-  along with ctak. If not, see <https://www.gnu.org/licenses/>.
-*/
-
-#include "nn.h"
-#include "tak.h"
-
-/*
- * Encode a board position into the network's input vector:
- *   [0]       ply parity (-1 / +1)
- *   [1]       white flats remaining, as a fraction of the total
- *   [2]       black flats remaining, as a fraction of the total
- *   [3 ..]    one scalar per cell for the top stone:
- *             0 empty, +-0.25 standing, +-0.5 flat, +-1 capstone,
- *             + white, - black
- *
- * `cur_board' must have room for 3 + board_size*board_size floats.
- */
-
-// The per-cell scalar depends only on the top stone's colour, variant
-// and occupancy, so it's a table lookup keyed on those three bits.
-static const float cell_scalar[16] = {
-    0.0f,  -0.5f, 0.0f, -0.25f, 0.0f, -1.0f, 0.0f, 0.0f, // black
-    0.0f,  +0.5f, 0.0f, +0.25f, 0.0f, +1.0f, 0.0f, 0.0f, // white
-};
-
-void nn_prepare_input(float *cur_board, tak_state_p state) {
-  const float inv_max_flats =
-      (state->board_size == 6) ? (1.0f / 30.0f) : (1.0f / 21.0f);
-  const unsigned int n_cells = state->board_size * state->board_size;
-  for (unsigned int loc = 0; loc < n_cells; loc++) {
-    const uint8_t d = state->celldat[loc];
-    cur_board[3 + loc] = cell_scalar[((state->colours[loc] & 1) << 3) |
-                                     ((d & STONE_MASK) << 1) |
-                                     ((d & NUM_MASK) != 0)];
-  }
-  // Add input of flat counts and ply parity
-  cur_board[0] = (state->ply & 1) ? 1 : -1;
-  cur_board[1] = (float)(state->white_count & 127) * inv_max_flats;
-  cur_board[2] = (float)(state->black_count & 127) * inv_max_flats;
-}