mirror of
https://github.com/sockspls/badfish
synced 2025-05-03 18:19:35 +00:00

This patch ports the efficiently updatable neural network (NNUE) evaluation to Stockfish. Both the NNUE and the classical evaluations are available, and can be used to assign a value to a position that is later used in alpha-beta (PVS) search to find the best move. The classical evaluation computes this value as a function of various chess concepts, handcrafted by experts, tested and tuned using fishtest. The NNUE evaluation computes this value with a neural network based on basic inputs. The network is optimized and trained on the evalutions of millions of positions at moderate search depth. The NNUE evaluation was first introduced in shogi, and ported to Stockfish afterward. It can be evaluated efficiently on CPUs, and exploits the fact that only parts of the neural network need to be updated after a typical chess move. [The nodchip repository](https://github.com/nodchip/Stockfish) provides additional tools to train and develop the NNUE networks. This patch is the result of contributions of various authors, from various communities, including: nodchip, ynasu87, yaneurao (initial port and NNUE authors), domschl, FireFather, rqs, xXH4CKST3RXx, tttak, zz4032, joergoster, mstembera, nguyenpham, erbsenzaehler, dorzechowski, and vondele. This new evaluation needed various changes to fishtest and the corresponding infrastructure, for which tomtor, ppigazzini, noobpwnftw, daylen, and vondele are gratefully acknowledged. The first networks have been provided by gekkehenker and sergiovieri, with the latter net (nn-97f742aaefcd.nnue) being the current default. The evaluation function can be selected at run time with the `Use NNUE` (true/false) UCI option, provided the `EvalFile` option points the the network file (depending on the GUI, with full path). The performance of the NNUE evaluation relative to the classical evaluation depends somewhat on the hardware, and is expected to improve quickly, but is currently on > 80 Elo on fishtest: 60000 @ 10+0.1 th 1 https://tests.stockfishchess.org/tests/view/5f28fe6ea5abc164f05e4c4c ELO: 92.77 +-2.1 (95%) LOS: 100.0% Total: 60000 W: 24193 L: 8543 D: 27264 Ptnml(0-2): 609, 3850, 9708, 10948, 4885 40000 @ 20+0.2 th 8 https://tests.stockfishchess.org/tests/view/5f290229a5abc164f05e4c58 ELO: 89.47 +-2.0 (95%) LOS: 100.0% Total: 40000 W: 12756 L: 2677 D: 24567 Ptnml(0-2): 74, 1583, 8550, 7776, 2017 At the same time, the impact on the classical evaluation remains minimal, causing no significant regression: sprt @ 10+0.1 th 1 https://tests.stockfishchess.org/tests/view/5f2906a2a5abc164f05e4c5b LLR: 2.94 (-2.94,2.94) {-6.00,-4.00} Total: 34936 W: 6502 L: 6825 D: 21609 Ptnml(0-2): 571, 4082, 8434, 3861, 520 sprt @ 60+0.6 th 1 https://tests.stockfishchess.org/tests/view/5f2906cfa5abc164f05e4c5d LLR: 2.93 (-2.94,2.94) {-6.00,-4.00} Total: 10088 W: 1232 L: 1265 D: 7591 Ptnml(0-2): 49, 914, 3170, 843, 68 The needed networks can be found at https://tests.stockfishchess.org/nns It is recommended to use the default one as indicated by the `EvalFile` UCI option. Guidelines for testing new nets can be found at https://github.com/glinscott/fishtest/wiki/Creating-my-first-test#nnue-net-tests Integration has been discussed in various issues: https://github.com/official-stockfish/Stockfish/issues/2823 https://github.com/official-stockfish/Stockfish/issues/2728 The integration branch will be closed after the merge: https://github.com/official-stockfish/Stockfish/pull/2825 https://github.com/official-stockfish/Stockfish/tree/nnue-player-wip closes https://github.com/official-stockfish/Stockfish/pull/2912 This will be an exciting time for computer chess, looking forward to seeing the evolution of this approach. Bench: 4746616
371 lines
12 KiB
C++
371 lines
12 KiB
C++
/*
|
|
Stockfish, a UCI chess playing engine derived from Glaurung 2.1
|
|
Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
|
|
|
|
Stockfish 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.
|
|
|
|
Stockfish 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 this program. If not, see <http://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
#include <cassert>
|
|
#include <cmath>
|
|
#include <iostream>
|
|
#include <sstream>
|
|
#include <string>
|
|
|
|
#include "evaluate.h"
|
|
#include "movegen.h"
|
|
#include "position.h"
|
|
#include "search.h"
|
|
#include "thread.h"
|
|
#include "timeman.h"
|
|
#include "tt.h"
|
|
#include "uci.h"
|
|
#include "syzygy/tbprobe.h"
|
|
|
|
using namespace std;
|
|
|
|
extern vector<string> setup_bench(const Position&, istream&);
|
|
|
|
namespace {
|
|
|
|
// FEN string of the initial position, normal chess
|
|
const char* StartFEN = "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1";
|
|
|
|
|
|
// position() is called when engine receives the "position" UCI command.
|
|
// The function sets up the position described in the given FEN string ("fen")
|
|
// or the starting position ("startpos") and then makes the moves given in the
|
|
// following move list ("moves").
|
|
|
|
void position(Position& pos, istringstream& is, StateListPtr& states) {
|
|
|
|
Move m;
|
|
string token, fen;
|
|
|
|
is >> token;
|
|
|
|
if (token == "startpos")
|
|
{
|
|
fen = StartFEN;
|
|
is >> token; // Consume "moves" token if any
|
|
}
|
|
else if (token == "fen")
|
|
while (is >> token && token != "moves")
|
|
fen += token + " ";
|
|
else
|
|
return;
|
|
|
|
states = StateListPtr(new std::deque<StateInfo>(1)); // Drop old and create a new one
|
|
pos.set(fen, Options["UCI_Chess960"], &states->back(), Threads.main());
|
|
|
|
// Parse move list (if any)
|
|
while (is >> token && (m = UCI::to_move(pos, token)) != MOVE_NONE)
|
|
{
|
|
states->emplace_back();
|
|
pos.do_move(m, states->back());
|
|
}
|
|
}
|
|
|
|
// trace_eval() prints the evaluation for the current position, consistent with the UCI
|
|
// options set so far.
|
|
|
|
void trace_eval(Position& pos) {
|
|
|
|
StateListPtr states(new std::deque<StateInfo>(1));
|
|
Position p;
|
|
p.set(pos.fen(), Options["UCI_Chess960"], &states->back(), Threads.main());
|
|
|
|
Eval::verify_NNUE();
|
|
|
|
sync_cout << "\n" << Eval::trace(p) << sync_endl;
|
|
}
|
|
|
|
|
|
// setoption() is called when engine receives the "setoption" UCI command. The
|
|
// function updates the UCI option ("name") to the given value ("value").
|
|
|
|
void setoption(istringstream& is) {
|
|
|
|
string token, name, value;
|
|
|
|
is >> token; // Consume "name" token
|
|
|
|
// Read option name (can contain spaces)
|
|
while (is >> token && token != "value")
|
|
name += (name.empty() ? "" : " ") + token;
|
|
|
|
// Read option value (can contain spaces)
|
|
while (is >> token)
|
|
value += (value.empty() ? "" : " ") + token;
|
|
|
|
if (Options.count(name))
|
|
Options[name] = value;
|
|
else
|
|
sync_cout << "No such option: " << name << sync_endl;
|
|
}
|
|
|
|
|
|
// go() is called when engine receives the "go" UCI command. The function sets
|
|
// the thinking time and other parameters from the input string, then starts
|
|
// the search.
|
|
|
|
void go(Position& pos, istringstream& is, StateListPtr& states) {
|
|
|
|
Search::LimitsType limits;
|
|
string token;
|
|
bool ponderMode = false;
|
|
|
|
limits.startTime = now(); // As early as possible!
|
|
|
|
while (is >> token)
|
|
if (token == "searchmoves") // Needs to be the last command on the line
|
|
while (is >> token)
|
|
limits.searchmoves.push_back(UCI::to_move(pos, token));
|
|
|
|
else if (token == "wtime") is >> limits.time[WHITE];
|
|
else if (token == "btime") is >> limits.time[BLACK];
|
|
else if (token == "winc") is >> limits.inc[WHITE];
|
|
else if (token == "binc") is >> limits.inc[BLACK];
|
|
else if (token == "movestogo") is >> limits.movestogo;
|
|
else if (token == "depth") is >> limits.depth;
|
|
else if (token == "nodes") is >> limits.nodes;
|
|
else if (token == "movetime") is >> limits.movetime;
|
|
else if (token == "mate") is >> limits.mate;
|
|
else if (token == "perft") is >> limits.perft;
|
|
else if (token == "infinite") limits.infinite = 1;
|
|
else if (token == "ponder") ponderMode = true;
|
|
|
|
Threads.start_thinking(pos, states, limits, ponderMode);
|
|
}
|
|
|
|
|
|
// bench() is called when engine receives the "bench" command. Firstly
|
|
// a list of UCI commands is setup according to bench parameters, then
|
|
// it is run one by one printing a summary at the end.
|
|
|
|
void bench(Position& pos, istream& args, StateListPtr& states) {
|
|
|
|
string token;
|
|
uint64_t num, nodes = 0, cnt = 1;
|
|
|
|
vector<string> list = setup_bench(pos, args);
|
|
num = count_if(list.begin(), list.end(), [](string s) { return s.find("go ") == 0 || s.find("eval") == 0; });
|
|
|
|
TimePoint elapsed = now();
|
|
|
|
for (const auto& cmd : list)
|
|
{
|
|
istringstream is(cmd);
|
|
is >> skipws >> token;
|
|
|
|
if (token == "go" || token == "eval")
|
|
{
|
|
cerr << "\nPosition: " << cnt++ << '/' << num << endl;
|
|
if (token == "go")
|
|
{
|
|
go(pos, is, states);
|
|
Threads.main()->wait_for_search_finished();
|
|
nodes += Threads.nodes_searched();
|
|
}
|
|
else
|
|
trace_eval(pos);
|
|
}
|
|
else if (token == "setoption") setoption(is);
|
|
else if (token == "position") position(pos, is, states);
|
|
else if (token == "ucinewgame") { Search::clear(); elapsed = now(); } // Search::clear() may take some while
|
|
}
|
|
|
|
elapsed = now() - elapsed + 1; // Ensure positivity to avoid a 'divide by zero'
|
|
|
|
dbg_print(); // Just before exiting
|
|
|
|
cerr << "\n==========================="
|
|
<< "\nTotal time (ms) : " << elapsed
|
|
<< "\nNodes searched : " << nodes
|
|
<< "\nNodes/second : " << 1000 * nodes / elapsed << endl;
|
|
}
|
|
|
|
// The win rate model returns the probability (per mille) of winning given an eval
|
|
// and a game-ply. The model fits rather accurately the LTC fishtest statistics.
|
|
int win_rate_model(Value v, int ply) {
|
|
|
|
// The model captures only up to 240 plies, so limit input (and rescale)
|
|
double m = std::min(240, ply) / 64.0;
|
|
|
|
// Coefficients of a 3rd order polynomial fit based on fishtest data
|
|
// for two parameters needed to transform eval to the argument of a
|
|
// logistic function.
|
|
double as[] = {-8.24404295, 64.23892342, -95.73056462, 153.86478679};
|
|
double bs[] = {-3.37154371, 28.44489198, -56.67657741, 72.05858751};
|
|
double a = (((as[0] * m + as[1]) * m + as[2]) * m) + as[3];
|
|
double b = (((bs[0] * m + bs[1]) * m + bs[2]) * m) + bs[3];
|
|
|
|
// Transform eval to centipawns with limited range
|
|
double x = Utility::clamp(double(100 * v) / PawnValueEg, -1000.0, 1000.0);
|
|
|
|
// Return win rate in per mille (rounded to nearest)
|
|
return int(0.5 + 1000 / (1 + std::exp((a - x) / b)));
|
|
}
|
|
|
|
} // namespace
|
|
|
|
|
|
/// UCI::loop() waits for a command from stdin, parses it and calls the appropriate
|
|
/// function. Also intercepts EOF from stdin to ensure gracefully exiting if the
|
|
/// GUI dies unexpectedly. When called with some command line arguments, e.g. to
|
|
/// run 'bench', once the command is executed the function returns immediately.
|
|
/// In addition to the UCI ones, also some additional debug commands are supported.
|
|
|
|
void UCI::loop(int argc, char* argv[]) {
|
|
|
|
Position pos;
|
|
string token, cmd;
|
|
StateListPtr states(new std::deque<StateInfo>(1));
|
|
|
|
pos.set(StartFEN, false, &states->back(), Threads.main());
|
|
|
|
for (int i = 1; i < argc; ++i)
|
|
cmd += std::string(argv[i]) + " ";
|
|
|
|
do {
|
|
if (argc == 1 && !getline(cin, cmd)) // Block here waiting for input or EOF
|
|
cmd = "quit";
|
|
|
|
istringstream is(cmd);
|
|
|
|
token.clear(); // Avoid a stale if getline() returns empty or blank line
|
|
is >> skipws >> token;
|
|
|
|
if ( token == "quit"
|
|
|| token == "stop")
|
|
Threads.stop = true;
|
|
|
|
// The GUI sends 'ponderhit' to tell us the user has played the expected move.
|
|
// So 'ponderhit' will be sent if we were told to ponder on the same move the
|
|
// user has played. We should continue searching but switch from pondering to
|
|
// normal search.
|
|
else if (token == "ponderhit")
|
|
Threads.main()->ponder = false; // Switch to normal search
|
|
|
|
else if (token == "uci")
|
|
sync_cout << "id name " << engine_info(true)
|
|
<< "\n" << Options
|
|
<< "\nuciok" << sync_endl;
|
|
|
|
else if (token == "setoption") setoption(is);
|
|
else if (token == "go") go(pos, is, states);
|
|
else if (token == "position") position(pos, is, states);
|
|
else if (token == "ucinewgame") Search::clear();
|
|
else if (token == "isready") sync_cout << "readyok" << sync_endl;
|
|
|
|
// Additional custom non-UCI commands, mainly for debugging.
|
|
// Do not use these commands during a search!
|
|
else if (token == "flip") pos.flip();
|
|
else if (token == "bench") bench(pos, is, states);
|
|
else if (token == "d") sync_cout << pos << sync_endl;
|
|
else if (token == "eval") trace_eval(pos);
|
|
else if (token == "compiler") sync_cout << compiler_info() << sync_endl;
|
|
else
|
|
sync_cout << "Unknown command: " << cmd << sync_endl;
|
|
|
|
} while (token != "quit" && argc == 1); // Command line args are one-shot
|
|
}
|
|
|
|
|
|
/// UCI::value() converts a Value to a string suitable for use with the UCI
|
|
/// protocol specification:
|
|
///
|
|
/// cp <x> The score from the engine's point of view in centipawns.
|
|
/// mate <y> Mate in y moves, not plies. If the engine is getting mated
|
|
/// use negative values for y.
|
|
|
|
string UCI::value(Value v) {
|
|
|
|
assert(-VALUE_INFINITE < v && v < VALUE_INFINITE);
|
|
|
|
stringstream ss;
|
|
|
|
if (abs(v) < VALUE_MATE_IN_MAX_PLY)
|
|
ss << "cp " << v * 100 / PawnValueEg;
|
|
else
|
|
ss << "mate " << (v > 0 ? VALUE_MATE - v + 1 : -VALUE_MATE - v) / 2;
|
|
|
|
return ss.str();
|
|
}
|
|
|
|
|
|
/// UCI::wdl() report WDL statistics given an evaluation and a game ply, based on
|
|
/// data gathered for fishtest LTC games.
|
|
|
|
string UCI::wdl(Value v, int ply) {
|
|
|
|
stringstream ss;
|
|
|
|
int wdl_w = win_rate_model( v, ply);
|
|
int wdl_l = win_rate_model(-v, ply);
|
|
int wdl_d = 1000 - wdl_w - wdl_l;
|
|
ss << " wdl " << wdl_w << " " << wdl_d << " " << wdl_l;
|
|
|
|
return ss.str();
|
|
}
|
|
|
|
|
|
/// UCI::square() converts a Square to a string in algebraic notation (g1, a7, etc.)
|
|
|
|
std::string UCI::square(Square s) {
|
|
return std::string{ char('a' + file_of(s)), char('1' + rank_of(s)) };
|
|
}
|
|
|
|
|
|
/// UCI::move() converts a Move to a string in coordinate notation (g1f3, a7a8q).
|
|
/// The only special case is castling, where we print in the e1g1 notation in
|
|
/// normal chess mode, and in e1h1 notation in chess960 mode. Internally all
|
|
/// castling moves are always encoded as 'king captures rook'.
|
|
|
|
string UCI::move(Move m, bool chess960) {
|
|
|
|
Square from = from_sq(m);
|
|
Square to = to_sq(m);
|
|
|
|
if (m == MOVE_NONE)
|
|
return "(none)";
|
|
|
|
if (m == MOVE_NULL)
|
|
return "0000";
|
|
|
|
if (type_of(m) == CASTLING && !chess960)
|
|
to = make_square(to > from ? FILE_G : FILE_C, rank_of(from));
|
|
|
|
string move = UCI::square(from) + UCI::square(to);
|
|
|
|
if (type_of(m) == PROMOTION)
|
|
move += " pnbrqk"[promotion_type(m)];
|
|
|
|
return move;
|
|
}
|
|
|
|
|
|
/// UCI::to_move() converts a string representing a move in coordinate notation
|
|
/// (g1f3, a7a8q) to the corresponding legal Move, if any.
|
|
|
|
Move UCI::to_move(const Position& pos, string& str) {
|
|
|
|
if (str.length() == 5) // Junior could send promotion piece in uppercase
|
|
str[4] = char(tolower(str[4]));
|
|
|
|
for (const auto& m : MoveList<LEGAL>(pos))
|
|
if (str == UCI::move(m, pos.is_chess960()))
|
|
return m;
|
|
|
|
return MOVE_NONE;
|
|
}
|