1
0
Fork 0
mirror of https://github.com/sockspls/badfish synced 2025-05-02 01:29:36 +00:00
BadFish/src/tune.h
nodchip 84f3e86790 Add NNUE evaluation
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
2020-08-06 16:37:45 +02:00

193 lines
7.2 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/>.
*/
#ifndef TUNE_H_INCLUDED
#define TUNE_H_INCLUDED
#include <memory>
#include <string>
#include <type_traits>
#include <vector>
typedef std::pair<int, int> Range; // Option's min-max values
typedef Range (RangeFun) (int);
// Default Range function, to calculate Option's min-max values
inline Range default_range(int v) {
return v > 0 ? Range(0, 2 * v) : Range(2 * v, 0);
}
struct SetRange {
explicit SetRange(RangeFun f) : fun(f) {}
SetRange(int min, int max) : fun(nullptr), range(min, max) {}
Range operator()(int v) const { return fun ? fun(v) : range; }
RangeFun* fun;
Range range;
};
#define SetDefaultRange SetRange(default_range)
/// BoolConditions struct is used to tune boolean conditions in the
/// code by toggling them on/off according to a probability that
/// depends on the value of a tuned integer parameter: for high
/// values of the parameter condition is always disabled, for low
/// values is always enabled, otherwise it is enabled with a given
/// probability that depnends on the parameter under tuning.
struct BoolConditions {
void init(size_t size) { values.resize(size, defaultValue), binary.resize(size, 0); }
void set();
std::vector<int> binary, values;
int defaultValue = 465, variance = 40, threshold = 500;
SetRange range = SetRange(0, 1000);
};
extern BoolConditions Conditions;
inline void set_conditions() { Conditions.set(); }
/// Tune class implements the 'magic' code that makes the setup of a fishtest
/// tuning session as easy as it can be. Mainly you have just to remove const
/// qualifiers from the variables you want to tune and flag them for tuning, so
/// if you have:
///
/// const Score myScore = S(10, 15);
/// const Value myValue[][2] = { { V(100), V(20) }, { V(7), V(78) } };
///
/// If you have a my_post_update() function to run after values have been updated,
/// and a my_range() function to set custom Option's min-max values, then you just
/// remove the 'const' qualifiers and write somewhere below in the file:
///
/// TUNE(SetRange(my_range), myScore, myValue, my_post_update);
///
/// You can also set the range directly, and restore the default at the end
///
/// TUNE(SetRange(-100, 100), myScore, SetDefaultRange);
///
/// In case update function is slow and you have many parameters, you can add:
///
/// UPDATE_ON_LAST();
///
/// And the values update, including post update function call, will be done only
/// once, after the engine receives the last UCI option, that is the one defined
/// and created as the last one, so the GUI should send the options in the same
/// order in which have been defined.
class Tune {
typedef void (PostUpdate) (); // Post-update function
Tune() { read_results(); }
Tune(const Tune&) = delete;
void operator=(const Tune&) = delete;
void read_results();
static Tune& instance() { static Tune t; return t; } // Singleton
// Use polymorphism to accomodate Entry of different types in the same vector
struct EntryBase {
virtual ~EntryBase() = default;
virtual void init_option() = 0;
virtual void read_option() = 0;
};
template<typename T>
struct Entry : public EntryBase {
static_assert(!std::is_const<T>::value, "Parameter cannot be const!");
static_assert( std::is_same<T, int>::value
|| std::is_same<T, Value>::value
|| std::is_same<T, Score>::value
|| std::is_same<T, PostUpdate>::value, "Parameter type not supported!");
Entry(const std::string& n, T& v, const SetRange& r) : name(n), value(v), range(r) {}
void operator=(const Entry&) = delete; // Because 'value' is a reference
void init_option() override;
void read_option() override;
std::string name;
T& value;
SetRange range;
};
// Our facilty to fill the container, each Entry corresponds to a parameter to tune.
// We use variadic templates to deal with an unspecified number of entries, each one
// of a possible different type.
static std::string next(std::string& names, bool pop = true);
int add(const SetRange&, std::string&&) { return 0; }
template<typename T, typename... Args>
int add(const SetRange& range, std::string&& names, T& value, Args&&... args) {
list.push_back(std::unique_ptr<EntryBase>(new Entry<T>(next(names), value, range)));
return add(range, std::move(names), args...);
}
// Template specialization for arrays: recursively handle multi-dimensional arrays
template<typename T, size_t N, typename... Args>
int add(const SetRange& range, std::string&& names, T (&value)[N], Args&&... args) {
for (size_t i = 0; i < N; i++)
add(range, next(names, i == N - 1) + "[" + std::to_string(i) + "]", value[i]);
return add(range, std::move(names), args...);
}
// Template specialization for SetRange
template<typename... Args>
int add(const SetRange&, std::string&& names, SetRange& value, Args&&... args) {
return add(value, (next(names), std::move(names)), args...);
}
// Template specialization for BoolConditions
template<typename... Args>
int add(const SetRange& range, std::string&& names, BoolConditions& cond, Args&&... args) {
for (size_t size = cond.values.size(), i = 0; i < size; i++)
add(cond.range, next(names, i == size - 1) + "_" + std::to_string(i), cond.values[i]);
return add(range, std::move(names), args...);
}
std::vector<std::unique_ptr<EntryBase>> list;
public:
template<typename... Args>
static int add(const std::string& names, Args&&... args) {
return instance().add(SetDefaultRange, names.substr(1, names.size() - 2), args...); // Remove trailing parenthesis
}
static void init() { for (auto& e : instance().list) e->init_option(); read_options(); } // Deferred, due to UCI::Options access
static void read_options() { for (auto& e : instance().list) e->read_option(); }
static bool update_on_last;
};
// Some macro magic :-) we define a dummy int variable that compiler initializes calling Tune::add()
#define STRINGIFY(x) #x
#define UNIQUE2(x, y) x ## y
#define UNIQUE(x, y) UNIQUE2(x, y) // Two indirection levels to expand __LINE__
#define TUNE(...) int UNIQUE(p, __LINE__) = Tune::add(STRINGIFY((__VA_ARGS__)), __VA_ARGS__)
#define UPDATE_ON_LAST() bool UNIQUE(p, __LINE__) = Tune::update_on_last = true
// Some macro to tune toggling of boolean conditions
#define CONDITION(x) (Conditions.binary[__COUNTER__] || (x))
#define TUNE_CONDITIONS() int UNIQUE(c, __LINE__) = (Conditions.init(__COUNTER__), 0); \
TUNE(Conditions, set_conditions)
#endif // #ifndef TUNE_H_INCLUDED