Trying minimax

1 files changed, 258 insertions, 0 deletions

diff --git a/include/ct1986.c b/include/ct1986.c
new file mode 100644
index 0000000..7295088
--- /dev/null
+++ b/include/ct1986.c
@@ -0,0 +1,258 @@
+#include "ct1986.h"
+
+// ===================================================================
+// Globals
+// ===================================================================
+
+char ct1986_ptn[9];
+void (*ct1986_display_progress)(const uint8_t);
+
+// ===================================================================
+// Implementation of a small convolutional neural network
+// ===================================================================
+
+static float flattened[CONV_NUM+2];
+static float dense1[DENSE1_NUM];
+static float dense2[DENSE2_NUM];
+
+#define RELU(x) ((x) = ((x)<0)?0:(x))
+
+float
+ct1986_evaluate_black_win(void) {
+	/* ------------------ *
+	 *  Convolution layer *
+	 * ------------------ */
+	// for each kernel
+	for (uint8_t kern = 0; kern < KERN_NUM; kern++) {
+		// the stride is 1, march across the board
+		for (uint8_t bx = 0; bx < KERN_OSIZE; bx++) {
+			for (uint8_t by = 0; by < KERN_OSIZE; by++) {
+				flattened[kern+KERN_NUM*(bx+by*KERN_OSIZE)] =
+					conv2d_biases[kern];
+				// Compute the convolution for this position
+				for (uint8_t ky = 0; ky < KERN_SIZE; ky++) {
+					for (uint8_t kx = 0; kx < KERN_SIZE; kx++) {
+						for (uint8_t c = 0; c < KERN_CHAN; c++) {
+							// Where we are on the board
+							const uint8_t loc = kx+by+(ky+bx)*board_size;
+							// Look up what's on the board at this location, and
+							// multiply it. For c=0 we have to do some extra work
+							float lookup = 0;
+							if (COUNT_AT(loc)>c) {
+								if (c==0) {
+									if (STONE_AT(loc) == STONE_STANDING) {
+										lookup = (colours[loc] & 1) ? +0.25 : -0.25;
+									} else if (STONE_AT(loc) == STONE_CAPSTONE) {
+										lookup = (colours[loc] & 1) ? +1.00 : -1.00;
+									} else {
+										lookup = (colours[loc] & 1) ? +0.50 : -0.50;
+									}
+								} else {
+									lookup = (colours[loc] & (1<<c)) ? +0.50 : -0.50;
+								}
+							}
+							flattened[kern+KERN_NUM*(bx+by*KERN_OSIZE)]
+								+= lookup*conv2d_weights[kern][ky][kx][c];
+						}
+					}
+				}
+				RELU(flattened[kern+KERN_NUM*(bx+by*KERN_OSIZE)]);
+			}
+		}
+	}
+	// Add input of flat counts
+	flattened[CONV_NUM]   = (float)(white_count & 127)/21.0;
+	flattened[CONV_NUM+1] = (float)(black_count & 127)/21.0;
+	/* ------------------ *
+	 *  First dense layer *
+	 * ------------------ */
+	for (uint8_t d1 = 0; d1 < DENSE1_NUM; d1++) {
+		dense1[d1] = dense1_biases[d1];
+		for (uint8_t fl = 0; fl < CONV_NUM+2; fl++) {
+			dense1[d1] += flattened[fl]*dense1_weights[d1][fl];
+		}
+		RELU(dense1[d1]);
+	}
+	/* ------------------- *
+	 *  Second dense layer *
+	 * ------------------- */
+	for (uint8_t d2 = 0; d2 < DENSE2_NUM; d2++) {
+		dense2[d2] = dense2_biases[d2];
+		for (uint8_t d1 = 0; d1 < DENSE1_NUM; d1++) {
+			dense2[d2] += dense1[d1]*dense2_weights[d2][d1];
+		}
+		RELU(dense2[d2]);
+	}
+	/* ------------- *
+	 *  Output layer *
+	 * ------------- */
+	float output = output_bias;
+	for (uint8_t d2 = 0; d2 < DENSE2_NUM; d2++) {
+		output += dense2[d2]*output_weights[d2];
+	}
+
+	// Truncated Pade approximant of logistic function
+	output = (12.0+output+50.0*output/(output*output+10.0))/24.0;
+	/*
+	 * // Truncated Pade approximant of tanh
+	 * output = output/6+25*output/(6*(2*output*output+5));
+	 */
+	if (output > 1.0) return 1.0;
+	else if (output < 0.0) return 0.0;
+
+	return output;
+}
+
+// ===================================================================
+// Minimax using the above evaluator
+// ===================================================================
+
+static void
+previous_ply(void) {
+	if (ply>0) ply--;
+	if (ply == 1) {
+		current_colour = C_WHITE;
+	} else {
+		if (current_colour == C_BLACK) current_colour = C_WHITE;
+		else current_colour = C_BLACK;
+	}
+}
+
+float
+ct1986_minimax(const uint8_t cur_depth, const uint8_t max_depth,
+							 const uint8_t min) {
+	enum E_RESULT r;
+	uint16_t colours_backup[board_size];
+	uint8_t celldat_backup[board_size], drops[board_size];
+	const uint8_t white_count_backup = white_count,
+		black_count_backup = black_count;
+
+	// 1.0 is a `certain' black win, -1.0 is a `certain' white win.
+	float this = 0, optimal = (min) ? 2.0 : -2.0;
+
+	// Step across the board
+	for (uint8_t row = 0; row < board_size; row++) {
+		for (uint8_t col = 0; col < board_size; col++) {
+			// Try all valid actions for this square. Is it empty?
+			const uint8_t loc = THE_COORDS(col, row);
+			const uint8_t count = COUNT_AT(loc);
+			// Only try moves after CPS
+			if (count && ((colours[loc] & 1) == current_colour) && ply>2) {
+				// There are stones, can we move them in a given direction?
+				// I'm not a huge fan of looping through enums, but it's
+				// better than manually unrolling this. Sufficiently smart
+				// compilers?
+
+				// Back up the row of the board
+				for (uint8_t y = 0; y < board_size; y++) {
+					colours_backup[y] = colours[THE_COORDS(col, y)];
+					celldat_backup[y] = celldat[THE_COORDS(col, y)];
+				}
+				for (enum MOVE_DIRECTION dir = M_UP; dir <= M_RIGHT; dir++) {
+					// Back-up the column once we start looking horizontally
+					if (dir == M_LEFT) {
+						for (uint8_t x = 0; x < board_size; x++) {
+							colours_backup[x] = colours[THE_COORDS(x, row)];
+							celldat_backup[x] = celldat[THE_COORDS(x, row)];
+						}
+					}
+					// We don't do anything terribly efficient or smart here,
+					// just try everything...
+
+					// For every number of steps
+					for (uint8_t steps = 1; steps < board_size && steps <= count; steps++) {
+						uint8_t idx, carry;
+						for (idx = 0; idx < steps; idx++) drops[idx]=0;
+						idx = 0;
+						while (idx < steps) {
+							// Increment the drop sequence
+							carry = 0;
+							drops[idx]++;
+							do {
+								if (carry) { drops[++idx]++; carry = 0;}
+								if (drops[idx] > count || drops[idx] > board_size) {
+									drops[idx] = 1; carry = 1;
+								}
+							} while (carry && idx < steps);
+							// If carry is still set here we're done
+							if (carry == 0) {
+								// Try it, and note that try_move will never return
+								// GAME_END. It does not check for winners. We don't
+								// presently do that either, trust in the magic
+								// numbers :)
+								r = try_move(loc, dir, steps, drops);
+								if (r == ACT_OK) {
+									// Decide what to do based on our depth
+									if (cur_depth == max_depth) {
+										// We're at the bottom, evaluate
+										this = ct1986_evaluate_black_win();
+									} else {
+										// We're not at the bottom, recurse first
+										next_ply();
+										this = ct1986_minimax(cur_depth + 1, max_depth, 1-min);
+										previous_ply();
+									}
+									// Update depending on min and optimal
+									if ( (min && (this < optimal))
+											 || ((min==0) && (this > optimal))) {
+										optimal = this;
+										if (cur_depth == 0)
+											generate_move(loc, dir, steps, drops, ct1986_ptn);
+									}
+									// Reset the board data
+									if (dir <= M_DOWN) {
+										for (uint8_t y = 0; y < board_size; y++) {
+											colours[THE_COORDS(col, y)] = colours_backup[y];
+											celldat[THE_COORDS(col, y)] = celldat_backup[y];
+										}
+									} else {
+										for (uint8_t x = 0; x < board_size; x++) {
+											colours[THE_COORDS(x, row)] = colours_backup[x];
+											celldat[THE_COORDS(x, row)] = celldat_backup[x];
+										}
+									}
+								}
+							}
+						}
+					}
+				}
+			} else if (count == 0) {
+				// Empty square, try the three placements. Again, looping
+				// through enums, sigh.
+				for (enum STONE_VARIANT stone = STONE_FLAT;
+						 stone <= STONE_CAPSTONE; stone++) {
+					// try_place will never check for winning, and we don't do
+					// that either here
+					r = try_place(loc, current_colour, stone);
+					// Legal placement, evaluate it
+					if (r == ACT_OK) {
+						// Decide what to do based on our depth
+						if (cur_depth == max_depth) {
+							// We're at the bottom, evaluate
+							this = ct1986_evaluate_black_win();
+						} else {
+							// We're not at the bottom, recurse first
+							next_ply();
+							this = ct1986_minimax(cur_depth + 1, max_depth, 1-min);
+							previous_ply();
+						}
+						// Update depending on min and optimal
+						if ( (min && (this < optimal))
+								 || ((min==0) && (this > optimal))) {
+							optimal = this;
+							// Store the result if we're at the top
+							if (cur_depth == 0)
+								generate_place(loc, stone, ct1986_ptn);
+						}
+						// Reset the state
+						celldat[loc] = 0;
+						white_count = white_count_backup;
+						black_count = black_count_backup;
+					}
+				}
+			}
+			ct1986_display_progress(cur_depth);
+		}
+	}
+	return optimal;
+}