This repository has been archived on 2023-08-20. You can view files and clone it, but cannot push or open issues or pull requests.
yap-6.3/packages/yap-lbfgs/lbfgs.pl

260 lines
9.5 KiB
Perl
Raw Normal View History

2013-06-13 23:57:55 +01:00
%%% -*- Mode: Prolog; -*-
2017-04-07 23:10:59 +01:00
%% @file lbfgs.pl
2013-06-13 23:57:55 +01:00
% This file is part of YAP-LBFGS.
% Copyright (C) 2009 Bernd Gutmann
%
% YAP-LBFGS 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.
%
% YAP-LBFGS 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 YAP-LBFGS. If not, see <http://www.gnu.org/licenses/>.
2019-04-04 13:27:06 +01:00
:- module(lbfgs,[lbfgs_initialize/2,
2018-10-05 10:26:34 +01:00
lbfgs_initialize/4,
2019-04-04 13:27:06 +01:00
lbfgs_run/3,
2013-06-13 23:57:55 +01:00
2018-10-05 10:26:34 +01:00
lbfgs_finalize/1,
2013-06-13 23:57:55 +01:00
2019-04-03 15:04:24 +01:00
lbfgs_set_parameter/2,
lbfgs_get_parameter/2,
lbfgs_parameters/0]).
2013-06-13 23:57:55 +01:00
% switch on all the checks to reduce bug searching time
% :- yap_flag(unknown,error).
% :- style_check(single_var).
2014-09-15 09:13:50 +01:00
/**
2017-04-07 23:10:59 +01:00
@defgroup YAP-LBFGS Interface to LibLBFGS
2015-01-04 23:58:23 +00:00
@ingroup packages
2014-09-15 09:13:50 +01:00
2017-04-07 23:10:59 +01:00
@short What is YAP-LBFGS? YAP-LBFGS is an interface to call [libLBFG](http://www.chokkan.org/software/liblbfgs/), from within
2014-09-15 09:13:50 +01:00
YAP. libLBFGS is a C library for Limited-memory
Broyden-Fletcher-Goldfarb-Shanno (L-BFGS) solving the under-constrained
minimization problem:
2017-04-07 23:10:59 +01:00
~~~~~~~~~~~~~~~~~~~~~~~~
2014-09-15 09:13:50 +01:00
+ minimize `F(X), X=(x1,x2,..., xN)`
2017-04-07 23:10:59 +01:00
~~~~~~~~~~~~~~~~~~~~~~~~
2014-09-15 09:13:50 +01:00
2018-10-05 10:26:34 +01:00
### Contact YAP-LBFGS has been developed by Bernd Gutmann. In case you
publish something using YAP-LBFGS, please give credit to me and to
libLBFGS. And if you find YAP-LBFGS useful, or if you find a bug, or
if you port it to another system, ... please send me an email.
2014-09-15 09:13:50 +01:00
### License
+ YAP-LBFGS 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.
+ YAP-LBFGS 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.
### Usage</h2>
The module lbfgs provides the following predicates after you loaded
2018-10-07 14:27:01 +01:00
it by
2014-09-15 09:13:50 +01:00
~~~~
:-use_module(library(lbfgs)).
~~~~
2018-10-05 10:26:34 +01:00
+ use lbfgs_set_paramater(Name,Value) to change parameters
+ use lbfgs_get_parameter(Name,Value) to see current parameters
+ use lbfgs_parameters to print this overview
2014-09-15 09:13:50 +01:00
### Demo
The following Prolog program, ex1.pl, searches for minimas of the
function `f(x0)=sin(x0)`. In order to do so, it provides the
call back predicate <span class="code">evaluate` which
calculates `f(x0)` and the gradient `d/dx0 f=cos(x0)`.
~~~~~
:- use_module(lbfgs).
% This is the call back function which evaluates F and the gradient of F
2018-10-05 10:26:34 +01:00
evaluate(FX,X,G,_N,_Step,_User) :-
X0 <== X[0],
F is sin(X0),
FX[0] <== F,
2014-09-15 09:13:50 +01:00
G0 is cos(X0),
2018-10-05 10:26:34 +01:00
G[0] <== G0.
2014-09-15 09:13:50 +01:00
% This is the call back function which is invoked to report the progress
2018-10-05 10:26:34 +01:00
% if the last argument is set to anything else than 0, the lbfgs will
2014-09-15 09:13:50 +01:00
% stop right now
2018-10-05 10:26:34 +01:00
progress(FX,X,X_Norm,G_Norm,Step,_N,Iteration,Ls,0) :-
X0 <== X[0],
2014-09-15 09:13:50 +01:00
format('~d. Iteration : x0=~4f f(X)=~4f |X|=~4f
|X\'|=~4f Step=~4f Ls=~4f~n',
[Iteration,X0,FX,X_Norm,G_Norm,Step,Ls]).
demo :-
format('Optimizing the function f(x0) = sin(x0)~n',[]),
2018-10-05 10:26:34 +01:00
lbfgs_initialize(1,X,0,Solver),
2014-09-15 09:13:50 +01:00
StartX is random*10,
format('We start the search at the random position x0=~5f~2n',[StartX]),
2018-10-05 10:26:34 +01:00
X[0] <== StartX,
2018-10-07 14:27:01 +01:00
2018-10-05 10:26:34 +01:00
lbfgs_run(Solver,BestF,Status),
BestX0 <== X[0],
lbfgs_finalize(Solver),
2014-09-15 09:13:50 +01:00
format('~2nOptimization done~nWe found a minimum at
f(~f)=~f~2nLBFGS Status=~w~n',[BestX0,BestF,Status]).
~~~~~
The output of this program is something like:
~~~~~
?- demo.
Optimizing the function f(x0) = sin(x0)
We start the search at the random position x0=7.24639
1. Iteration : x0=5.0167 f(X)=-0.9541 |X|=5.0167 |X'|=0.2996 Step=3.9057 Ls=3.0000
2. Iteration : x0=4.7708 f(X)=-0.9983 |X|=4.7708 |X'|=0.0584 Step=0.0998 Ls=2.0000
3. Iteration : x0=4.7113 f(X)=-1.0000 |X|=4.7113 |X'|=0.0011 Step=1.0000 Ls=1.0000
4. Iteration : x0=4.7124 f(X)=-1.0000 |X|=4.7124 |X'|=0.0000 Step=1.0000 Ls=1.0000
Optimization done
We found a minimum at f(4.712390)=-1.000000
LBFGS Status=0
yes
?-
~~~~~
@{
*/
2013-06-13 23:57:55 +01:00
2018-08-21 03:01:03 +01:00
:- load_foreign_files(['libLBFGS'],[],'init_lbfgs_predicates').
2013-06-13 23:57:55 +01:00
2018-10-05 10:26:34 +01:00
/** @pred lbfgs_initialize(+N, -SolverInfo)
2013-06-13 23:57:55 +01:00
2018-10-08 13:51:17 +01:00
Do initial memory allocation and a reference to a descriptor.
2014-09-15 09:13:50 +01:00
~~~~
2018-10-05 10:26:34 +01:00
lbfgs_initialize(1, Block)
2014-09-15 09:13:50 +01:00
~~~~~
*/
2019-04-04 13:27:06 +01:00
lbfgs_initialize(N,X) :-
2018-10-08 13:51:17 +01:00
integer(N),
N>0,
lbfgs_grab(N,X).
2013-06-13 23:57:55 +01:00
% install call back predicates in the user module which call
2018-10-07 14:27:01 +01:00
% the predicates given by the arguments
2013-06-13 23:57:55 +01:00
2018-10-05 10:26:34 +01:00
2018-10-08 13:51:17 +01:00
/** @pred lbfgs_finalize(+State)
2014-09-15 09:13:50 +01:00
Clean up the memory.
*/
2019-04-04 13:27:06 +01:00
lbfgs_finalize(_N).
2018-10-05 10:26:34 +01:00
2018-10-08 13:51:17 +01:00
/** @pred lbfgs_run(+State, -FinalOutput)
run the algorithm. output the final score of the function being optimised
2018-09-13 13:35:37 +01:00
*/
2019-04-04 13:27:06 +01:00
lbfgs_run(N,X,FX) :-
lbfgs(N,X, FX).
2018-09-13 13:35:37 +01:00
2013-06-13 23:57:55 +01:00
2014-09-15 09:13:50 +01:00
2019-04-03 15:04:24 +01:00
/** @pred lbfgs_parameters/0
2014-09-15 09:13:50 +01:00
Prints a table with the current parameters. See the <a href="http://www.chokkan.org/software/liblbfgs/structlbfgs__parameter__t.html#_details">documentation
of libLBFGS</a> for the meaning of each parameter.
~~~~
2018-10-15 13:48:49 +01:00
?- lbfgs_parameters(State).
2014-09-15 09:13:50 +01:00
==========================================================================================
2018-10-07 14:27:01 +01:00
Type Name Value Description
2014-09-15 09:13:50 +01:00
==========================================================================================
int m 6 The number of corrections to approximate the inverse hessian matrix.
2018-10-07 14:27:01 +01:00
float epsilon 1e-05 Epsilon for convergence test.
2014-09-15 09:13:50 +01:00
int past 0 Distance for delta-based convergence test.
2018-10-07 14:27:01 +01:00
float delta 1e-05 Delta for convergence test.
2014-09-15 09:13:50 +01:00
int max_iterations 0 The maximum number of iterations
2018-10-07 14:27:01 +01:00
int linesearch 0 The line search algorithm.
2014-09-15 09:13:50 +01:00
int max_linesearch 40 The maximum number of trials for the line search.
float min_step 1e-20 The minimum step of the line search routine.
float max_step 1e+20 The maximum step of the line search.
float ftol 0.0001 A parameter to control the accuracy of the line search routine.
float gtol 0.9 A parameter to control the accuracy of the line search routine.
float xtol 1e-16 The machine precision for floating-point values.
float orthantwise_c 0.0 Coefficient for the L1 norm of variables
int orthantwise_start 0 Start index for computing the L1 norm of the variables.
int orthantwise_end -1 End index for computing the L1 norm of the variables.
==========================================================================================
2018-10-07 14:27:01 +01:00
~~~~
2014-09-15 09:13:50 +01:00
*/
2018-10-05 10:26:34 +01:00
lbfgs_parameters :-
2019-04-03 15:04:24 +01:00
lbfgs_get_parameter(m,M ),
lbfgs_get_parameter(epsilon,Epsilon ),
lbfgs_get_parameter(past,Past ),
lbfgs_get_parameter(delta,Delta ),
lbfgs_get_parameter(max_iterations,Max_Iterations ),
lbfgs_get_parameter(linesearch,Linesearch ),
lbfgs_get_parameter(max_linesearch,Max_Linesearch ),
lbfgs_get_parameter(min_step,Min_Step ),
lbfgs_get_parameter(max_step,Max_Step ),
lbfgs_get_parameter(ftol,Ftol ),
lbfgs_get_parameter(gtol,Gtol ),
lbfgs_get_parameter(xtol,Xtol ),
lbfgs_get_parameter(orthantwise_c,Orthantwise_C ),
lbfgs_get_parameter(orthantwise_start,Orthantwise_Start ),
lbfgs_get_parameter(orthantwise_end,Orthantwise_End ),
2018-09-13 13:35:37 +01:00
format('/******************************************************************************************~n',[] ),
2019-04-03 15:04:24 +01:00
print_param('Name','Value','Description','Type' ),
2018-09-13 13:35:37 +01:00
format('******************************************************************************************~n',[] ),
2019-04-03 15:04:24 +01:00
print_param(m,M,'The number of corrections to approximate the inverse hessian matrix.',int ),
print_param(epsilon,Epsilon,'Epsilon for convergence test.',float ),
print_param(past,Past,'Distance for delta-based convergence test.',int ),
print_param(delta,Delta,'Delta for convergence test.',float ),
print_param(max_iterations,Max_Iterations,'The maximum number of iterations',int ),
print_param(linesearch,Linesearch,'The line search algorithm.',int ),
print_param(max_linesearch,Max_Linesearch,'The maximum number of trials for the line search.',int ),
print_param(min_step,Min_Step,'The minimum step of the line search routine.',float ),
print_param(max_step,Max_Step,'The maximum step of the line search.',float ),
print_param(ftol,Ftol,'A parameter to control the accuracy of the line search routine.',float ),
print_param(gtol,Gtol,'A parameter to control the accuracy of the line search routine.',float ),
print_param(xtol,Xtol,'The machine precision for floating-point values.',float ),
print_param(orthantwise_c,Orthantwise_C,'Coefficient for the L1 norm of variables',float ),
print_param(orthantwise_start,Orthantwise_Start,'Start index for computing the L1 norm of the variables.',int ),
print_param(orthantwise_end,Orthantwise_End,'End index for computing the L1 norm of the variables.',int ),
2013-07-29 23:55:51 +01:00
format('******************************************************************************************/~n',[]),
2019-04-03 15:04:24 +01:00
format(' use lbfgs_set_parameter(Name,Value) to change parameters~n',[]),
2019-04-04 13:27:06 +01:00
format(' use lbfgs_get_parameter(Name,Values) to see current parameters~n',[]),
2018-10-05 10:26:34 +01:00
format(' use lbfgs_parameters to print this overview~2n',[]).
2013-06-13 23:57:55 +01:00
print_param(Name,Value,Text,Dom) :-
format(user,'~w~10+~w~19+~w~15+~w~30+~n',[Dom,Name,Value,Text]).
2018-05-20 00:47:27 +01:00
%% @}