LBFGS
This commit is contained in:
Vitor Santos Costa
parent
724681dde8
commit
5ea98bcf53
@ -14,8 +14,8 @@
|
||||
% will run 20 iterations of learning with default settings
|
||||
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
||||
|
||||
:- use_module(library(problog)).
|
||||
:- use_module(library(problog_learning)).
|
||||
:- use_module(library(matrix)).
|
||||
:- use_module(('../problog_lbfgs')).
|
||||
|
||||
%%%%
|
||||
% background knowledge
|
||||
@ -99,3 +99,7 @@ test_example(33,path(5,4),0.57).
|
||||
test_example(34,path(6,4),0.51).
|
||||
test_example(35,path(6,5),0.69).
|
||||
|
||||
:- set_problog_flag(init_method,(Query,_,BDD,
|
||||
problog_exact_lbdd(user:Query,BDD))).
|
||||
|
||||
|
||||
|
||||
@ -6,24 +6,23 @@
|
||||
:- use_module(library(bdd)).
|
||||
:- use_module(library(bhash)).
|
||||
|
||||
problog_exact_lbdd(Goal,Prob,Status) :-
|
||||
problog_control(on, exact),
|
||||
problog_low_lbdd(Goal,0,Prob,Status),
|
||||
problog_control(off, exact).
|
||||
problog_exact_lbdd(Goal,BDD) :-
|
||||
problog_low_lbdd(Goal, 0, _, _, BDD).
|
||||
|
||||
problog_low_lbdd(Goal, Threshold, _, _) :-
|
||||
problog_low_lbdd(Goal, Threshold, _, _, _) :-
|
||||
init_problog_low(Threshold),
|
||||
problog_control(off, up),
|
||||
timer_start(sld_time),
|
||||
problog_call(Goal),
|
||||
add_solution,
|
||||
fail.
|
||||
problog_low_lbdd(_, _, Prob, ok) :-
|
||||
problog_low_lbdd(_, _, Prob, ok, bdd(Dir, Tree, MapList)) :-
|
||||
timer_stop(sld_time,SLD_Time),
|
||||
problog_var_set(sld_time, SLD_Time),
|
||||
nb_getval(problog_completed_proofs, Trie_Completed_Proofs),
|
||||
tabled_trie_to_bdd(Trie_Completed_Proofs, BDD, MapList),
|
||||
trie_to_bdd(Trie_Completed_Proofs, BDD, MapList),
|
||||
bind_maplist(MapList, BoundVars),
|
||||
bdd_tree(BDD, bdd(Dir, Tree, _Vars)),
|
||||
bdd_to_probability_sum_product(BDD, BoundVars, Prob),
|
||||
(problog_flag(verbose, true)->
|
||||
problog_statistics
|
||||
@ -73,6 +72,23 @@ problog_fl_bdd(_,Prob) :-
|
||||
(problog_flag(retain_tables, true) -> retain_tabling; true),
|
||||
clear_tabling.
|
||||
|
||||
problog_full_bdd(Goal,_K, _) :-
|
||||
init_problog_low(0.0),
|
||||
problog_control(off, up),
|
||||
timer_start(sld_time),
|
||||
problog_call(Goal),
|
||||
add_solution,
|
||||
fail.
|
||||
problog_full_bdd(_,Prob) :-
|
||||
timer_stop(sld_time,SLD_Time),
|
||||
problog_var_set(sld_time, SLD_Time),
|
||||
nb_getval(problog_completed_proofs, Trie_Completed_Proofs),
|
||||
tabled_trie_to_bdd(Trie_Completed_Proofs, BDD, MapList),
|
||||
bind_maplist(MapList, BoundVars),
|
||||
bdd_to_probability_sum_product(BDD, BoundVars, Prob),
|
||||
(problog_flag(retain_tables, true) -> retain_tabling; true),
|
||||
clear_tabling.
|
||||
|
||||
bind_maplist([], []).
|
||||
bind_maplist([Node-_|MapList], [ProbFact|BoundVars]) :-
|
||||
get_fact_probability(Node,ProbFact),
|
||||
|
||||
1012
packages/ProbLog/problog_lbfgs.yap
Normal file
1012
packages/ProbLog/problog_lbfgs.yap
Normal file
File diff suppressed because it is too large
Load Diff
@ -22,35 +22,36 @@
|
||||
|
||||
:- use_module(library(matrix)).
|
||||
|
||||
|
||||
|
||||
% This is the call back function which evaluates F and the gradient of F
|
||||
evaluate(FX,X,G,_N,_Step) :-
|
||||
evaluate(FX,X,G,_N,_Step,_User) :-
|
||||
X0 <== X[0],
|
||||
FX is sin(X0),
|
||||
G0 is cos(X0),
|
||||
G[0] <== G0.
|
||||
|
||||
% This is the call back function which is invoked to report the progress
|
||||
% if the last argument is set to anywhting else than 0, the optimizer will
|
||||
% if the last argument is set to anything else than 0, the lbfgs will
|
||||
% stop right now
|
||||
progress(FX,X,G,X_Norm,G_Norm,Step,_N,Iteration,Ls, 0) :-
|
||||
progress(FX,X,G,X_Norm,G_Norm,Step,_N,Iteration,Ls,0) :-
|
||||
X0 <== X[0],
|
||||
format('~d. Iteration : x0=~4f f(X)=~4f |X|=~4f |X\'|=~4f Step=~4f Ls=~4f~n',
|
||||
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',[]),
|
||||
optimizer_initialize(1,X,Status),
|
||||
lbfgs_initialize(1,X,0,Solver),
|
||||
|
||||
|
||||
StartX is random*10,
|
||||
format('We start the search at the random position x0=~5f~2n',[StartX]),
|
||||
X[0] <== StartX,
|
||||
|
||||
optimizer_run(Status, BestF, BestX0, O),
|
||||
lbfgs_run(Solver,BestF,Status),
|
||||
BestX0 <== X[0],
|
||||
optimizer_finalize(Status),
|
||||
format('~2nOptimization done~nWe found a minimum at f(~f)=~f~2nLBFGS Status=~w~n',[BestX0,BestF,O]).
|
||||
|
||||
|
||||
lbfgs_finalize(Solver),
|
||||
format('~2nOptimization done~nWe found a minimum at
|
||||
f(~f)=~f~2nLBFGS Status=~w~n',[BestX0,BestF,Status]).
|
||||
|
||||
@ -20,13 +20,14 @@
|
||||
:- use_module(library(lbfgs)).
|
||||
:- use_module(library(matrix)).
|
||||
|
||||
f(X0,X1,FX) :-
|
||||
FX is (X0-2)*(X0-2) + (X1-1)*(X1-1).
|
||||
|
||||
% This is the call back function which evaluates F and the gradient of F
|
||||
evaluate(FX,X,G,_N,_Step) :-
|
||||
evaluate(FX,X,G,_N,_Step,_U) :-
|
||||
X0 <== X[0],
|
||||
X1 <== X[1],
|
||||
|
||||
FX is (X0-2)*(X0-2) + (X1-1)*(X1-1),
|
||||
f(X0,X1,FX),
|
||||
G0 is 2*(X0-2),
|
||||
G1 is 2*(X1-2),
|
||||
G[0] <== G0,
|
||||
@ -44,7 +45,8 @@ progress(FX,X,_G,X_Norm,G_Norm,Step,_N,Iteration,Ls,0) :-
|
||||
|
||||
demo :-
|
||||
format('Optimizing the function f(x0,x1) = (x0-2)^2 + (x1-1)^2~n',[]),
|
||||
optimizer_initialize(2,X,Status),
|
||||
|
||||
lbfgs_initialize(2,X,0,Solver),
|
||||
|
||||
|
||||
StartX0 is random*1000-500,
|
||||
@ -53,13 +55,11 @@ demo :-
|
||||
format('We start the search at the random position (x0,x1)=(~5f,~5f)~2n',[StartX0,StartX1]),
|
||||
X[0] <== StartX0,
|
||||
X[1] <== StartX1,
|
||||
|
||||
|
||||
optimizer_run(Status,BestF,BestX0, O),
|
||||
lbfgs_run(Solver,BestF,Status),
|
||||
BestX0 <== X[0],
|
||||
BestX1 <== X[1],
|
||||
|
||||
optimizer_finalize,
|
||||
optimizer_finalize(Solver),
|
||||
format('~2nOptimization done~nWe found a minimum at f(~f,~f)=~f~2nLBFGS Status=~w~n',[BestX0,BestX1,BestF,Status]).
|
||||
|
||||
|
||||
|
||||
@ -20,14 +20,16 @@
|
||||
|
||||
|
||||
|
||||
:- module(lbfgs,[optimizer_initialize/3,
|
||||
optimizer_run/4,
|
||||
:- module(lbfgs,[lbfgs_initialize/3,
|
||||
lbfgs_initialize/4,
|
||||
lbfgs_run/2,
|
||||
|
||||
optimizer_finalize/1,
|
||||
lbfgs_finalize/1,
|
||||
|
||||
optimizer_set_parameter/3,
|
||||
optimizer_get_parameter/3,
|
||||
optimizer_parameters/1]).
|
||||
lbfgs_set_parameter/3,
|
||||
lbfgs_get_parameter/3,
|
||||
lbfgs_parameters/0,
|
||||
lbfgs_parameters/1]).
|
||||
|
||||
% switch on all the checks to reduce bug searching time
|
||||
% :- yap_flag(unknown,error).
|
||||
@ -48,9 +50,11 @@ minimization problem:
|
||||
~~~~~~~~~~~~~~~~~~~~~~~~
|
||||
|
||||
|
||||
### 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.
|
||||
### 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.
|
||||
|
||||
|
||||
|
||||
### License
|
||||
@ -72,9 +76,9 @@ it by
|
||||
:-use_module(library(lbfgs)).
|
||||
~~~~
|
||||
|
||||
+ use optimizer_set_paramater(Name,Value) to change parameters
|
||||
+ use optimizer_get_parameter(Name,Value) to see current parameters
|
||||
+ use optimizer_parameters to print this overview
|
||||
+ 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
|
||||
|
||||
|
||||
|
||||
@ -89,17 +93,18 @@ 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
|
||||
evaluate(FX,_N,_Step) :-
|
||||
optimizer_get_x(0,X0),
|
||||
FX is sin(X0),
|
||||
evaluate(FX,X,G,_N,_Step,_User) :-
|
||||
X0 <== X[0],
|
||||
F is sin(X0),
|
||||
FX[0] <== F,
|
||||
G0 is cos(X0),
|
||||
optimizer_set_g(0,G0).
|
||||
G[0] <== G0.
|
||||
|
||||
% This is the call back function which is invoked to report the progress
|
||||
% if the last argument is set to anything else than 0, the optimizer will
|
||||
% if the last argument is set to anything else than 0, the lbfgs will
|
||||
% stop right now
|
||||
progress(FX,X_Norm,G_Norm,Step,_N,Iteration,Ls,0) :-
|
||||
optimizer_get_x(0,X0),
|
||||
progress(FX,X,X_Norm,G_Norm,Step,_N,Iteration,Ls,0) :-
|
||||
X0 <== X[0],
|
||||
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]).
|
||||
@ -108,16 +113,16 @@ progress(FX,X_Norm,G_Norm,Step,_N,Iteration,Ls,0) :-
|
||||
|
||||
demo :-
|
||||
format('Optimizing the function f(x0) = sin(x0)~n',[]),
|
||||
optimizer_initialize(1,evaluate,progress),
|
||||
lbfgs_initialize(1,X,0,Solver),
|
||||
|
||||
|
||||
StartX is random*10,
|
||||
format('We start the search at the random position x0=~5f~2n',[StartX]),
|
||||
optimizer_set_x(0,StartX),
|
||||
X[0] <== StartX,
|
||||
|
||||
optimizer_run(BestF,Status),
|
||||
optimizer_get_x(0,BestX0),
|
||||
optimizer_finalize,
|
||||
lbfgs_run(Solver,BestF,Status),
|
||||
BestX0 <== X[0],
|
||||
lbfgs_finalize(Solver),
|
||||
format('~2nOptimization done~nWe found a minimum at
|
||||
f(~f)=~f~2nLBFGS Status=~w~n',[BestX0,BestF,Status]).
|
||||
~~~~~
|
||||
@ -151,49 +156,54 @@ yes
|
||||
:- load_foreign_files(['libLBFGS'],[],'init_lbfgs_predicates').
|
||||
|
||||
|
||||
/** @pred optimizer_initialize(+N,+Evaluate,+Progress)
|
||||
/** @pred lbfgs_initialize(+N, -SolverInfo)
|
||||
The same as before, except that the user module is the default
|
||||
value.
|
||||
|
||||
Example
|
||||
~~~~
|
||||
optimizer_initialize(1)
|
||||
lbfgs_initialize(1, Block)
|
||||
~~~~~
|
||||
*/
|
||||
lbfgs_initialize(N,X,t(N,X,U,Params)) :-
|
||||
lbfgs_initialize(N,X,0,t(N,X,U,Params)).
|
||||
|
||||
optimizer_initialize(N,X,t(N,X,XO,Params)) :-
|
||||
|
||||
lbfgs_initialize(N,X,U,t(N,X,U,Params)) :-
|
||||
lbfgs_defaults(Params),
|
||||
integer(N),
|
||||
N>0,
|
||||
|
||||
% check whether there are such call back functions
|
||||
|
||||
optimizer_reserve_memory(N,X,XO,Params).
|
||||
lbfgs_grab(N,X).
|
||||
|
||||
% install call back predicates in the user module which call
|
||||
% the predicates given by the arguments
|
||||
|
||||
/** @pred optimizer_finalize/0
|
||||
|
||||
/** @pred lbfgs_finalize/0
|
||||
Clean up the memory.
|
||||
*/
|
||||
optimizer_finalize(t(N,X,XO,Params)) :-
|
||||
lbfgs_finalize(t(N,X,U,Params)) :-
|
||||
initialized,
|
||||
optimizer_free_memory(X,XO,Params) ,
|
||||
lbfgs_release(X) ,
|
||||
lbfgs_release_parameters(Params) ,
|
||||
retractall(initialized).
|
||||
/** @pred optimizer_run/3
|
||||
|
||||
/** @pred lbfgs_run/2
|
||||
Do the work.
|
||||
*/
|
||||
optimizer_run(t(N,X,XO,Params),FX,XO,Status) :-
|
||||
optimizer_run(N,X, FX, XO, Status, Params).
|
||||
lbfgs_run(t(N,X,U,Params),FX) :-
|
||||
lbfgs(N,X, Params, U, FX).
|
||||
|
||||
|
||||
|
||||
/** @pred optimizer_parameters/1
|
||||
/** @pred lbfgs_parameters/1
|
||||
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.
|
||||
|
||||
~~~~
|
||||
?- optimizer_parameters.
|
||||
?- lbfgs_parameters.
|
||||
==========================================================================================
|
||||
Type Name Value Description
|
||||
==========================================================================================
|
||||
@ -215,22 +225,26 @@ int orthantwise_end -1 End index for computing the L1 norm
|
||||
==========================================================================================
|
||||
~~~~
|
||||
*/
|
||||
optimizer_parameterse(t(X,XO,Params)) :-
|
||||
optimizer_get_parameter(m,M ,Params),
|
||||
optimizer_get_parameter(epsilon,Epsilon ,Params),
|
||||
optimizer_get_parameter(past,Past ,Params),
|
||||
optimizer_get_parameter(delta,Delta ,Params),
|
||||
optimizer_get_parameter(max_iterations,Max_Iterations ,Params),
|
||||
optimizer_get_parameter(linesearch,Linesearch ,Params),
|
||||
optimizer_get_parameter(max_linesearch,Max_Linesearch ,Params),
|
||||
optimizer_get_parameter(min_step,Min_Step ,Params),
|
||||
optimizer_get_parameter(max_step,Max_Step ,Params),
|
||||
optimizer_get_parameter(ftol,Ftol ,Params),
|
||||
optimizer_get_parameter(gtol,Gtol ,Params),
|
||||
optimizer_get_parameter(xtol,Xtol ,Params),
|
||||
optimizer_get_parameter(orthantwise_c,Orthantwise_C ,Params),
|
||||
optimizer_get_parameter(orthantwise_start,Orthantwise_Start ,Params),
|
||||
optimizer_get_parameter(orthantwise_end,Orthantwise_End ,Params),
|
||||
lbfgs_parameters :-
|
||||
lbfgs_defaults(Params),
|
||||
lbfgs_parameters(t(_X,_,_,Params)).
|
||||
|
||||
lbfgs_parameters(t(_,_,_,Params)) :-
|
||||
lbfgs_get_parameter(m,M ,Params),
|
||||
lbfgs_get_parameter(epsilon,Epsilon ,Params),
|
||||
lbfgs_get_parameter(past,Past ,Params),
|
||||
lbfgs_get_parameter(delta,Delta ,Params),
|
||||
lbfgs_get_parameter(max_iterations,Max_Iterations ,Params),
|
||||
lbfgs_get_parameter(linesearch,Linesearch ,Params),
|
||||
lbfgs_get_parameter(max_linesearch,Max_Linesearch ,Params),
|
||||
lbfgs_get_parameter(min_step,Min_Step ,Params),
|
||||
lbfgs_get_parameter(max_step,Max_Step ,Params),
|
||||
lbfgs_get_parameter(ftol,Ftol ,Params),
|
||||
lbfgs_get_parameter(gtol,Gtol ,Params),
|
||||
lbfgs_get_parameter(xtol,Xtol ,Params),
|
||||
lbfgs_get_parameter(orthantwise_c,Orthantwise_C ,Params),
|
||||
lbfgs_get_parameter(orthantwise_start,Orthantwise_Start ,Params),
|
||||
lbfgs_get_parameter(orthantwise_end,Orthantwise_End ,Params),
|
||||
|
||||
format('/******************************************************************************************~n',[] ),
|
||||
print_param('Name','Value','Description','Type' ,Params),
|
||||
@ -251,9 +265,9 @@ optimizer_parameterse(t(X,XO,Params)) :-
|
||||
print_param(orthantwise_start,Orthantwise_Start,'Start index for computing the L1 norm of the variables.',int ,Params),
|
||||
print_param(orthantwise_end,Orthantwise_End,'End index for computing the L1 norm of the variables.',int ,Params),
|
||||
format('******************************************************************************************/~n',[]),
|
||||
format(' use optimizer_set_paramater(Name,Value) to change parameters~n',[]),
|
||||
format(' use optimizer_get_parameter(Name,Value) to see current parameters~n',[]),
|
||||
format(' use optimizer_parameters to print this overview~2n',[]).
|
||||
format(' use lbfgs_set_parameter(Name,Value,Solver) to change parameters~n',[]),
|
||||
format(' use lbfgs_get_parameter(Name,Value,Solver) to see current parameters~n',[]),
|
||||
format(' use lbfgs_parameters to print this overview~2n',[]).
|
||||
|
||||
|
||||
print_param(Name,Value,Text,Dom) :-
|
||||
|
||||
@ -24,11 +24,11 @@
|
||||
|
||||
|
||||
// These constants describe the internal state
|
||||
#define OPTIMIZER_STATUS_NONE 0
|
||||
#define OPTIMIZER_STATUS_INITIALIZED 1
|
||||
#define OPTIMIZER_STATUS_RUNNING 2
|
||||
#define OPTIMIZER_STATUS_CB_EVAL 3
|
||||
#define OPTIMIZER_STATUS_CB_PROGRESS 4
|
||||
#define LBFGS_STATUS_NONE 0
|
||||
#define LBFGS_STATUS_INITIALIZED 1
|
||||
#define LBFGS_STATUS_RUNNING 2
|
||||
#define LBFGS_STATUS_CB_EVAL 3
|
||||
#define LBFGS_STATUS_CB_PROGRESS 4
|
||||
|
||||
X_API void init_lbfgs_predicates( void ) ;
|
||||
|
||||
@ -44,53 +44,39 @@ static lbfgsfloatval_t evaluate(
|
||||
)
|
||||
{
|
||||
YAP_Term call;
|
||||
YAP_Term v, a1;
|
||||
YAP_Bool result;
|
||||
YAP_Int s1;
|
||||
lbfgsfloatval_t rc;
|
||||
|
||||
YAP_Term t[5], t2[2];
|
||||
YAP_Term t[6], t2[2];
|
||||
|
||||
t[0] = v = YAP_MkVarTerm();
|
||||
t[0] = YAP_MkIntTerm((YAP_Int)&rc);
|
||||
t[0] = YAP_MkApplTerm(ffloats, 1, t);
|
||||
t[1] = YAP_MkIntTerm((YAP_Int)x);
|
||||
t[1] = YAP_MkApplTerm(ffloats, 1, t+1);
|
||||
t[2] = YAP_MkIntTerm((YAP_Int)g_tmp);
|
||||
t[2] = YAP_MkApplTerm(ffloats, 1, t+2);
|
||||
t[3] = YAP_MkIntTerm(n);
|
||||
t[4] = YAP_MkFloatTerm(step);
|
||||
t[5] = YAP_MkIntTerm((YAP_Int)instance);
|
||||
|
||||
t2[0] = tuser;
|
||||
t2[1] = YAP_MkApplTerm(fevaluate, 5, t);
|
||||
t2[1] = YAP_MkApplTerm(fevaluate, 6, t);
|
||||
|
||||
|
||||
call = YAP_MkApplTerm( fmodule, 2, t2 );
|
||||
|
||||
|
||||
s1 = YAP_InitSlot(v);
|
||||
//optimizer_status=OPTIMIZER_STATUS_CB_EVAL;
|
||||
result=YAP_RunGoal(call);
|
||||
//optimizer_status=OPTIMIZER_STATUS_RUNNING;
|
||||
// s1 = YAP_InitSlot(v);
|
||||
//lbfgs_status=LBFGS_STATUS_CB_EVAL;
|
||||
result=YAP_RunGoalOnce(call);
|
||||
//lbfgs_status=LBFGS_STATUS_RUNNING;
|
||||
|
||||
if (result==FALSE) {
|
||||
printf("ERROR: the evaluate call failed in YAP.\n");
|
||||
// Goal did not succeed
|
||||
YAP_ShutdownGoal( false );
|
||||
return FALSE;
|
||||
}
|
||||
|
||||
a1 = YAP_GetFromSlot( s1 );
|
||||
|
||||
lbfgsfloatval_t rc;
|
||||
if (YAP_IsFloatTerm(a1)) {
|
||||
rc = (lbfgsfloatval_t) YAP_FloatOfTerm(a1);
|
||||
} else if (YAP_IsIntTerm(a1)) {
|
||||
rc = (lbfgsfloatval_t) YAP_IntOfTerm(a1);
|
||||
} else {
|
||||
fprintf(stderr, "ERROR: The evaluate call back function did not return a number as first argument.\n");
|
||||
rc = false;
|
||||
|
||||
}
|
||||
|
||||
YAP_ShutdownGoal( false );
|
||||
return rc;
|
||||
}
|
||||
|
||||
@ -132,12 +118,11 @@ static int progress(
|
||||
call = YAP_MkApplTerm( fmodule, 2, t2 );
|
||||
s1 = YAP_InitSlot(v);
|
||||
|
||||
//optimizer_status=OPTIMIZER_STATUS_CB_PROGRESS;
|
||||
result=YAP_RunGoal(call);
|
||||
//optimizer_status=OPTIMIZER_STATUS_RUNNING;
|
||||
//lbfgs_status=LBFGS_STATUS_CB_PROGRESS;
|
||||
result=YAP_RunGoalOnce(call);
|
||||
//lbfgs_status=LBFGS_STATUS_RUNNING;
|
||||
|
||||
YAP_Term o = YAP_GetFromSlot( s1 );
|
||||
YAP_ShutdownGoal( false );
|
||||
|
||||
if (result==FALSE) {
|
||||
printf("ERROR: the progress call failed in YAP.\n");
|
||||
@ -154,7 +139,7 @@ static int progress(
|
||||
return 1;
|
||||
}
|
||||
|
||||
/** @pred optimizer_initialize(+N,+Module,+Evaluate,+Progress)
|
||||
/** @pred lbfgs_initialize(+N,+Module,+Evaluate,+Progress)
|
||||
Create space to optimize a function with _N_ variables (_N_ has to be
|
||||
integer).
|
||||
|
||||
@ -169,7 +154,7 @@ to evaluate the function math <span class="math">_F</span>_,
|
||||
|
||||
Example
|
||||
~~~~
|
||||
optimizer_initialize(1,user,evaluate,progress,e,g)</span>
|
||||
lbfgs_initialize(1,user,evaluate,progress,e,g)</span>
|
||||
~~~~
|
||||
|
||||
|
||||
@ -179,107 +164,227 @@ value _F_. _N_ is the
|
||||
size of the parameter vector (the value which was used to initialize
|
||||
LBFGS) and _Step_ is the current state of the
|
||||
line search. The call back predicate can access the current values of
|
||||
`x[i]` by calling `optimizer_get_x(+I,-Xi)`. Finally, the call back
|
||||
`x[i]` by calling `lbfgs_get_x(+I,-Xi)`. Finally, the call back
|
||||
predicate has to calculate the gradient of _F</span>_
|
||||
and set its value by calling `optimizer_set_g(+I,+Gi)` for every `1<=I<=N`.
|
||||
and set its value by calling `lbfgs_set_g(+I,+Gi)` for every `1<=I<=N`.
|
||||
|
||||
|
||||
The progress call back predicate has to be of the type
|
||||
`progress(+F,+X_Norm,+G_Norm,+Step,+N,+Iteration,+LS,-Continue)`. It
|
||||
is called after every iteration. The call back predicate can access
|
||||
the current values of _X_ and of the gradient by calling
|
||||
`optimizer_get_x(+I,-Xi)` and `optimizer_get_g`(+I,-Gi)`
|
||||
`lbfgs_get_x(+I,-Xi)` and `lbfgs_get_g`(+I,-Gi)`
|
||||
respectively. However, it must not call the setter predicates for <span
|
||||
class="code"_X_ or _G_. If it tries to do so, the optimizer will
|
||||
class="code"_X_ or _G_. If it tries to do so, the lbfgs will
|
||||
terminate with an error. If _Continue_ is set to 0 (int) the
|
||||
optimization process will continue for one more iteration, any other
|
||||
value will terminate the optimization process.
|
||||
*/
|
||||
static YAP_Bool optimizer_initialize(void) {
|
||||
/**
|
||||
* @pred lbfgs( N, X, U, FX )
|
||||
*
|
||||
* @Arg1 N: number of variables in problem
|
||||
* @Arg[X0]: input vector
|
||||
* @Arg[FX]: function value,
|
||||
* @Arg[FX]: parameter
|
||||
* @Arg[X0]: user data
|
||||
* @Arg[FX]: status
|
||||
*/
|
||||
static YAP_Bool p_lbfgs(void)
|
||||
{
|
||||
YAP_Term t1 = YAP_ARG1;
|
||||
int temp_n=0;
|
||||
lbfgsfloatval_t *temp_x, *temp_ox;
|
||||
lbfgs_parameter_t *temp_p;
|
||||
|
||||
int n;
|
||||
lbfgsfloatval_t *x;
|
||||
lbfgsfloatval_t fx;
|
||||
|
||||
if (! YAP_IsIntTerm(t1)) {
|
||||
return false;
|
||||
}
|
||||
|
||||
temp_n=YAP_IntOfTerm(t1);
|
||||
n=YAP_IntOfTerm(t1);
|
||||
|
||||
if (temp_n<1) {
|
||||
if (n<1) {
|
||||
return FALSE;
|
||||
}
|
||||
x = ( lbfgsfloatval_t*) YAP_IntOfTerm(YAP_ArgOfTerm(1,YAP_ARG2));
|
||||
lbfgs_parameter_t* param = ( lbfgs_parameter_t*) YAP_IntOfTerm(YAP_ARG3);
|
||||
void* ui = ( void*) YAP_IntOfTerm(YAP_ARG4);
|
||||
{
|
||||
YAP_Term t = YAP_MkIntTerm((YAP_Int)&fx);
|
||||
if (!YAP_Unify(YAP_ARG5,YAP_MkApplTerm(ffloats,1,&t)))
|
||||
return false;
|
||||
}
|
||||
signal(SIGFPE, SIG_IGN);
|
||||
int ret = lbfgs(n, x, &fx, evaluate, progress, ui, param);
|
||||
if (ret == 0)
|
||||
return true;
|
||||
const char *s;
|
||||
switch(ret) {
|
||||
case LBFGS_CONVERGENCE:
|
||||
case LBFGS_STOP:
|
||||
return true;
|
||||
/** The initial variables already minimize the objective function. */
|
||||
case LBFGS_ALREADY_MINIMIZED:
|
||||
s = "The initial variables already minimize the objective function.";
|
||||
break;
|
||||
case LBFGSERR_UNKNOWNERROR:
|
||||
s = "Unknownerror";
|
||||
break;
|
||||
case LBFGSERR_LOGICERROR:
|
||||
s = "logic error.";
|
||||
break;
|
||||
|
||||
temp_n = 16*(temp_n/16+15);
|
||||
lbfgs_parameter_init((temp_p=(lbfgs_parameter_t *)malloc(sizeof(lbfgs_parameter_t))));
|
||||
temp_ox = lbfgs_malloc(temp_n);
|
||||
YAP_Term tox = YAP_MkIntTerm((YAP_Int)temp_ox);
|
||||
temp_x = lbfgs_malloc(temp_n);
|
||||
YAP_Term tx = YAP_MkIntTerm((YAP_Int)temp_x);
|
||||
tx = YAP_MkApplTerm(ffloats, 1, &tx);
|
||||
tox = YAP_MkApplTerm(ffloats, 1, &tox);
|
||||
YAP_Term tp = YAP_MkIntTerm((YAP_Int)temp_p);
|
||||
case LBFGSERR_OUTOFMEMORY:
|
||||
s = "out of memory";
|
||||
break;
|
||||
case LBFGSERR_CANCELED:
|
||||
s = "canceled.";
|
||||
break;
|
||||
case LBFGSERR_INVALID_N:
|
||||
s = "Invalid number of variables specified.";
|
||||
break;
|
||||
|
||||
case LBFGSERR_INVALID_N_SSE:
|
||||
s = "Invalid number of variables (for SSE) specified.";
|
||||
break;
|
||||
|
||||
return YAP_Unify(YAP_ARG2,tx) && YAP_Unify(YAP_ARG3,tox) && YAP_Unify(YAP_ARG4,tp) ;
|
||||
case LBFGSERR_INVALID_X_SSE:
|
||||
s = "The array x must be aligned to 16 (for SSE).";
|
||||
break;
|
||||
case LBFGSERR_INVALID_EPSILON:
|
||||
s = "Invalid parameter lbfgs_parameter_t::epsilon specified.";
|
||||
break;
|
||||
case LBFGSERR_INVALID_TESTPERIOD:
|
||||
s = "Invalid parameter lbfgs_parameter_t::past specified.";
|
||||
break;
|
||||
case LBFGSERR_INVALID_DELTA:
|
||||
s = "Invalid parameter lbfgs_parameter_t::delta specified.";
|
||||
break;
|
||||
case LBFGSERR_INVALID_LINESEARCH:
|
||||
s = "Invalid parameter lbfgs_parameter_t::linesearch specified.";
|
||||
break;
|
||||
case LBFGSERR_INVALID_MINSTEP:
|
||||
s = "Invalid parameter lbfgs_parameter_t::max_step specified.";
|
||||
break;
|
||||
case LBFGSERR_INVALID_MAXSTEP:
|
||||
s = "Invalid parameter lbfgs_parameter_t::max_step specified.";
|
||||
break;
|
||||
case LBFGSERR_INVALID_FTOL:
|
||||
s = "Invalid parameter lbfgs_parameter_t::ftol specified.";
|
||||
break;
|
||||
case LBFGSERR_INVALID_WOLFE:
|
||||
s = "Invalid parameter lbfgs_parameter_t::wolfe specified.";
|
||||
break;
|
||||
case LBFGSERR_INVALID_GTOL:
|
||||
s = "Invalid parameter lbfgs_parameter_t::gtol specified.";
|
||||
break;
|
||||
case LBFGSERR_INVALID_XTOL:
|
||||
s = "Invalid parameter lbfgs_parameter_t::xtol specified.";
|
||||
break;
|
||||
case LBFGSERR_INVALID_MAXLINESEARCH:
|
||||
s = "Invalid parameter lbfgs_parameter_t::max_linesearch specified.";
|
||||
break;
|
||||
case LBFGSERR_INVALID_ORTHANTWISE:
|
||||
s = "Invalid parameter lbfgs_parameter_t::orthantwise_c specified.";
|
||||
break;
|
||||
case LBFGSERR_INVALID_ORTHANTWISE_START:
|
||||
s = "Invalid parameter lbfgs_parameter_t::orthantwise_start specified.";
|
||||
break;
|
||||
case LBFGSERR_INVALID_ORTHANTWISE_END:
|
||||
s = "Invalid parameter lbfgs_parameter_t::orthantwise_end specified.";
|
||||
break;
|
||||
case LBFGSERR_OUTOFINTERVAL:
|
||||
s = "The line-search step went out of the interval of uncertainty.";
|
||||
break;
|
||||
|
||||
case LBFGSERR_INCORRECT_TMINMAX:
|
||||
s = "A logic error occurred; alternatively, the interval of uncertaity became too small.";
|
||||
break;
|
||||
case LBFGSERR_ROUNDING_ERROR:
|
||||
s = "A rounding error occurred; alternatively, no line-search s";
|
||||
break;
|
||||
case LBFGSERR_MINIMUMSTEP:
|
||||
s = "The line-search step became smaller than lbfgs_parameter_t::min_step.";
|
||||
break;
|
||||
case LBFGSERR_MAXIMUMSTEP:
|
||||
s = "The line-search step became larger than lbfgs_parameter_t::max_step.";
|
||||
break;
|
||||
case LBFGSERR_MAXIMUMLINESEARCH:
|
||||
s = "The line-search routine reaches the maximum number of evaluations.";
|
||||
break;
|
||||
case LBFGSERR_MAXIMUMITERATION:
|
||||
s = "The algorithm routine reaches the maximum number of iterations lbfgs_parameter_t::xtol.";
|
||||
break;
|
||||
case LBFGSERR_WIDTHTOOSMALL:
|
||||
s = "Relative width of the interval of uncertainty is at m";
|
||||
break;
|
||||
case LBFGSERR_INVALIDPARAMETERS:
|
||||
s = "A logic error (negative line-search step) occurred.";
|
||||
break;
|
||||
|
||||
}
|
||||
fprintf(stderr, "optimization terminated with code %d: %s\n", ret, s);
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
|
||||
|
||||
/** @pred optimizer_run(-F,-Status)
|
||||
Runs the optimization, _F is the best (minimal) function value and
|
||||
Status (int) is the status code returned by libLBFGS. Anything except
|
||||
0 indicates an error, see the documentation of libLBFGS for the
|
||||
meaning.
|
||||
*/
|
||||
static YAP_Bool optimizer_run(void) {
|
||||
int ret = 0;
|
||||
int n = YAP_IntOfTerm(YAP_ARG1);
|
||||
YAP_Int s1, s2;
|
||||
lbfgsfloatval_t fx;
|
||||
lbfgsfloatval_t *temp_x = ( lbfgsfloatval_t *)YAP_IntOfTerm( YAP_ArgOfTerm(1, YAP_ARG2)),
|
||||
*temp_ox = ( lbfgsfloatval_t *) YAP_IntOfTerm(YAP_ArgOfTerm(1,YAP_ARG4));
|
||||
lbfgs_parameter_t *temp_p = (lbfgs_parameter_t * ) YAP_IntOfTerm(YAP_ARG6);
|
||||
ret = lbfgs(n, temp_x, &fx, evaluate, progress, temp_ox, temp_p);
|
||||
static YAP_Bool lbfgs_grab (void)
|
||||
{
|
||||
int n=YAP_IntOfTerm(YAP_ARG1);
|
||||
|
||||
return YAP_Unify(YAP_MkIntTerm(ret), YAP_ARG5) &&
|
||||
YAP_Unify(YAP_MkFloatTerm(fx), YAP_ARG3);
|
||||
}
|
||||
if (n<1) {
|
||||
return FALSE;
|
||||
}
|
||||
lbfgsfloatval_t * x = lbfgs_malloc(n);
|
||||
YAP_Term t = YAP_MkIntTerm((YAP_Int)x);
|
||||
return YAP_Unify(YAP_ARG2,YAP_MkApplTerm(ffloats,1,&t));
|
||||
|
||||
}
|
||||
|
||||
|
||||
|
||||
static YAP_Bool optimizer_finalize( void ) {
|
||||
/* if (optimizer_status == OPTIMIZER_STATUS_NONE) { */
|
||||
/* printf("Error: Optimizer is not initialized.\n"); */
|
||||
static YAP_Bool lbfgs_parameters( void )
|
||||
{
|
||||
lbfgs_parameter_t *x = malloc(sizeof(lbfgs_parameter_t));
|
||||
lbfgs_parameter_init(x);
|
||||
return YAP_Unify(YAP_ARG1,YAP_MkIntTerm((YAP_Int)x));
|
||||
|
||||
}
|
||||
|
||||
static YAP_Bool lbfgs_release_parameters( void )
|
||||
{
|
||||
free((void *)YAP_IntOfTerm(YAP_ARG1));
|
||||
return true;
|
||||
|
||||
}
|
||||
|
||||
static YAP_Bool lbfgs_release( void ) {
|
||||
/* if (lbfgs_status == LBFGS_STATUS_NONE) { */
|
||||
/* printf("Error: Lbfgs is not initialized.\n"); */
|
||||
/* return FALSE; */
|
||||
/* } */
|
||||
|
||||
/* if (optimizer_status == OPTIMIZER_STATUS_INITIALIZED) { */
|
||||
lbfgs_free((void *)YAP_IntOfTerm(YAP_ArgOfTerm(1,YAP_ARG1)));
|
||||
lbfgs_free((void *)YAP_IntOfTerm(YAP_ArgOfTerm(1,YAP_ARG2)));
|
||||
lbfgs_free((void *)YAP_IntOfTerm(YAP_ARG3));
|
||||
/* if (lbfgs_status == LBFGS_STATUS_INITIALIZED) { */
|
||||
lbfgs_free((lbfgsfloatval_t *)YAP_IntOfTerm(YAP_ArgOfTerm(1,(YAP_ARG1))));
|
||||
|
||||
return TRUE;
|
||||
/* } */
|
||||
|
||||
/* printf("ERROR: Optimizer is running right now. Please wait till it is finished.\n"); */
|
||||
/* return FALSE; */
|
||||
}
|
||||
|
||||
|
||||
|
||||
/** @pred optimizer_set_parameter(+Name,+Value,+Parameters)
|
||||
Set the parameter Name to Value. Only possible while the optimizer
|
||||
/** @pred lbfgs_set_parameter(+Name,+Value,+Parameters)
|
||||
Set the parameter Name to Value. Only possible while the lbfgs
|
||||
is not running.
|
||||
*/
|
||||
static YAP_Bool optimizer_set_parameter( void ) {
|
||||
static YAP_Bool lbfgs_set_parameter( void ) {
|
||||
YAP_Term t1 = YAP_ARG1;
|
||||
YAP_Term t2 = YAP_ARG2;
|
||||
lbfgs_parameter_t *param = (lbfgs_parameter_t *) YAP_IntOfTerm(YAP_ARG3);
|
||||
/* if (optimizer_status != OPTIMIZER_STATUS_NONE && optimizer_status != OPTIMIZER_STATUS_INITIALIZED){ */
|
||||
/* printf("ERROR: Optimizer is running right now. Please wait till it is finished.\n"); */
|
||||
/* if (lbfgs_status != LBFGS_STATUS_NONE && lbfgs_status != LBFGS_STATUS_INITIALIZED){ */
|
||||
/* printf("ERROR: Lbfgs is running right now. Please wait till it is finished.\n"); */
|
||||
/* return FALSE; */
|
||||
/* } */
|
||||
|
||||
@ -430,11 +535,11 @@ static YAP_Bool optimizer_set_parameter( void ) {
|
||||
}
|
||||
|
||||
|
||||
/** @pred optimizer_get_parameter(+Name,-Value)</h3>
|
||||
/** @pred lbfgs_get_parameter(+Name,-Value)</h3>
|
||||
Get the current Value for Name
|
||||
*/
|
||||
|
||||
static YAP_Bool optimizer_get_parameter( void ) {
|
||||
static YAP_Bool lbfgs_get_parameter( void ) {
|
||||
YAP_Term t1 = YAP_ARG1;
|
||||
YAP_Term t2 = YAP_ARG2;
|
||||
lbfgs_parameter_t *param = (lbfgs_parameter_t *) YAP_IntOfTerm(YAP_ARG3);
|
||||
@ -487,7 +592,7 @@ static YAP_Bool optimizer_get_parameter( void ) {
|
||||
|
||||
X_API void init_lbfgs_predicates( void )
|
||||
{
|
||||
fevaluate = YAP_MkFunctor(YAP_LookupAtom("evaluate"), 5);
|
||||
fevaluate = YAP_MkFunctor(YAP_LookupAtom("evaluate"), 6);
|
||||
fprogress = YAP_MkFunctor(YAP_LookupAtom("progress"), 10);
|
||||
fmodule = YAP_MkFunctor(YAP_LookupAtom(":"), 2);
|
||||
ffloats = YAP_MkFunctor(YAP_LookupAtom("floats"), 1);
|
||||
@ -497,10 +602,12 @@ X_API void init_lbfgs_predicates( void )
|
||||
// lbfgs_parameter_init(¶m);
|
||||
|
||||
|
||||
YAP_UserCPredicate("optimizer_reserve_memory",optimizer_initialize,4);
|
||||
YAP_UserCPredicate("optimizer_run",optimizer_run,6);
|
||||
YAP_UserCPredicate("optimizer_free_memory",optimizer_finalize,3);
|
||||
YAP_UserCPredicate("lbfgs_grab",lbfgs_grab,2);
|
||||
YAP_UserCPredicate("lbfgs",p_lbfgs, 5);
|
||||
YAP_UserCPredicate("lbfgs_release",lbfgs_release,1);
|
||||
|
||||
YAP_UserCPredicate("optimizer_set_parameter",optimizer_set_parameter,3);
|
||||
YAP_UserCPredicate("optimizer_get_parameter",optimizer_get_parameter,3);
|
||||
YAP_UserCPredicate("lbfgs_defaults",lbfgs_parameters,1);
|
||||
YAP_UserCPredicate("lbfgs_release_parameters",lbfgs_release_parameters,1);
|
||||
YAP_UserCPredicate("lbfgs_set_parameter",lbfgs_set_parameter,3);
|
||||
YAP_UserCPredicate("lbfgs_get_parameter",lbfgs_get_parameter,3);
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user