void rateclass_t::learn(
double & lg_L,
double & aicc,
vector< pair< double, double > > & params,
const int nrestart
) const
{
params.clear();
params.push_back( make_pair( 1.0, 0.5 ) );
lg_L = EM( data, params );
aicc = _aicc( 1, lg_L, data.size() / factor );
for ( int i = 2; ; ++i ) {
double old_lg_L, old_aicc;
vector< pair< double, double > > old_params = params;
old_params.push_back( make_pair( 1.0, 0.5 ) );
initialize_params( old_params, 0 );
old_lg_L = EM( data, old_params );
for ( int j = 1; j < nrestart; ++j ) {
double new_lg_L;
vector< pair< double, double > > new_params = old_params;
initialize_params( new_params, j );
new_lg_L = EM( data, new_params );
if ( new_lg_L > old_lg_L ) {
old_lg_L = new_lg_L;
old_params = new_params;
}
}
old_aicc = _aicc( 2 * i, old_lg_L, data.size() / factor );
// if our AICc doesn't improve, we're done
if ( old_aicc >= aicc )
break;
aicc = old_aicc;
lg_L = old_lg_L;
params = old_params;
}
// we've actually rates corresponding to the majority,
// but we really want the inverse
for ( unsigned i = 0; i < params.size(); ++i )
params[ i ].second = 1.0 - params[ i ].second;
sort( params.begin(), params.end(), rate_cmp );
}
static VALUE
initialize(VALUE self) {
volatile VALUE vparams;
struct cparse_params *v;
vparams = Data_Make_Struct(CparseParams, struct cparse_params,
cparse_params_mark, -1, v);
rb_iv_set(self, "@vparams", vparams);
rb_ivar_set(self, id_yydebug, Qnil);
rb_ivar_set(self, id_action_table, Qnil);
rb_ivar_set(self, id_action_check, Qnil);
rb_ivar_set(self, id_action_default, Qnil);
rb_ivar_set(self, id_action_pointer, Qnil);
rb_ivar_set(self, id_goto_table, Qnil);
rb_ivar_set(self, id_goto_check, Qnil);
rb_ivar_set(self, id_goto_default, Qnil);
rb_ivar_set(self, id_goto_pointer, Qnil);
rb_ivar_set(self, id_nt_base, Qnil);
rb_ivar_set(self, id_reduce_table, Qnil);
rb_ivar_set(self, id_token_table, Qnil);
rb_ivar_set(self, id_shift_n, Qnil);
rb_ivar_set(self, id_reduce_n, Qnil);
rb_ivar_set(self, id_use_result, Qnil);
D_puts("starting cparse");
v->sys_debug = Qtrue;
vparams = initialize_params(vparams, self, Qnil, Qnil);
v->lex_is_iterator = Qfalse;
return self;
}
static VALUE
racc_cparse(VALUE parser)
{
volatile VALUE vparams;
struct cparse_params *v;
vparams = Data_Make_Struct(CparseParams, struct cparse_params,
cparse_params_mark, -1, v);
D_puts("starting cparse");
v->sys_debug = Qtrue;
vparams = initialize_params(vparams, parser, Qnil, Qnil);
v->lex_is_iterator = Qfalse;
parse_main(v, Qnil, Qnil, 0);
return v->retval;
}
static VALUE
racc_cparse(VALUE parser, VALUE arg, VALUE sysdebug)
{
volatile VALUE vparams;
struct cparse_params *v;
vparams = Data_Make_Struct(CparseParams, struct cparse_params,
cparse_params_mark, -1, v);
D_puts("starting cparse");
v->sys_debug = RTEST(sysdebug);
vparams = initialize_params(vparams, parser, arg, Qnil, Qnil);
v->lex_is_iterator = FALSE;
parse_main(v, Qnil, Qnil, 0);
return v->retval;
}
static VALUE
racc_yyparse(VALUE parser, VALUE lexer, VALUE lexmid)
{
volatile VALUE vparams;
struct cparse_params *v;
vparams = Data_Make_Struct(CparseParams, struct cparse_params,
cparse_params_mark, -1, v);
v->sys_debug = Qfalse;
D_puts("start C yyparse");
vparams = initialize_params(vparams, parser, lexer, lexmid);
v->lex_is_iterator = Qtrue;
D_puts("params initialized");
parse_main(v, Qnil, Qnil, 0);
call_lexer(v);
if (!v->fin) {
rb_raise(rb_eArgError, "%s() is finished before EndOfToken",
rb_id2name(v->lexmid));
}
return v->retval;
}
void parse_args(int argc, char *argv[])
{
long l;
int i;
const char *progname = argv[0];
struct headers *header;
initialize_params();
while ((i = getopt(argc, argv, opts)) > 0) {
switch (i) {
case 'H':
header = parse_header_param(optarg);
if (header == NULL) {
fprintf(stderr, "error parsing -H parameter\n");
print_help(stderr, progname);
exit(-1);
} else {
overwrite_header(&config_opts.headers, header);
}
break;
case 'C':
{
/* split on : if present
* first part is config_opts.connect
* second part is config_opts.connect_port
*/
char *portstr;
config_opts.connect = strdup(optarg);
portstr = strchr(config_opts.connect, ':');
if (portstr)
*portstr++ = '\0'; /* terminate the host part */
if (portstr && *portstr) {
errno = 0;
l = strtol(portstr, (char **)NULL, 10);
if (errno) {
perror("invalid connect port (-C) value");
print_help(stderr, progname);
exit(-1);
} else if (l < 0 || l >= USHRT_MAX) {
fprintf(stderr, "invalid connect port (-C): %ld\n",
l);
print_help(stderr, progname);
exit(-1);
}
config_opts.connect_port = (unsigned short)l;
}
}
break;
case 'c':
errno = 0;
l = strtol(optarg, (char **)NULL, 10);
if (errno) {
perror("invalid concurrency (-c) value");
print_help(stderr, progname);
exit(-1);
} else if (l <= 0 || l == LONG_MAX) {
fprintf(stderr, "invalid concurrency (-c): %ld\n", l);
print_help(stderr, progname);
exit(-1);
}
config_opts.concurrency = (int)l;
break;
case 'n':
errno = 0;
l = strtol(optarg, (char **)NULL, 10);
if (errno) {
perror("invalid request count (-n) value");
print_help(stderr, progname);
exit(-1);
} else if (l <= 0 || l == LONG_MAX) {
fprintf(stderr, "invalid request count (-n): %ld\n", l);
print_help(stderr, progname);
exit(-1);
}
config_opts.count = (int)l;
break;
case 'o':
config_opts.halfopen = 1;
break;
case 'v':
config_opts.verbose++;
break;
default:
fprintf(stderr, "Unknown option: %o\n", i);
case 'h':
print_help(stderr, progname);
exit(-1);
break;
}
}
argc -= optind;
argv += optind;
if (argc == 0) {
fprintf(stderr, "At least one URL required...");
print_help(stderr, progname);
exit(-1);
} else for (i = 0; i < argc; i++) {
struct urls *url = malloc(sizeof(*url));
RING_INIT(url);
url->url = strdup(argv[i]);
RING_APPEND(config_opts.urls, url);
}
}
int run_em(EM_ENG_PTR em_ptr)
{
int r, iterate, old_valid, converged, saved = 0;
double likelihood, log_prior;
double lambda, old_lambda = 0.0;
config_em(em_ptr);
for (r = 0; r < num_restart; r++) {
SHOW_PROGRESS_HEAD("#em-iters", r);
initialize_params();
itemp = daem ? itemp_init : 1.0;
iterate = 0;
/* [21 Aug 2007, by yuizumi]
* while-loop for inversed temperature (DAEM). Note that this
* loop is evaluated only once for EM without annealing, since
* itemp initially set to 1.0 by the code above.
*/
while (1) {
if (daem) {
SHOW_PROGRESS_TEMP(itemp);
}
old_valid = 0;
while (1) {
if (CTRLC_PRESSED) {
SHOW_PROGRESS_INTR();
RET_ERR(err_ctrl_c_pressed);
}
RET_ON_ERR(em_ptr->compute_inside());
RET_ON_ERR(em_ptr->examine_inside());
likelihood = em_ptr->compute_likelihood();
log_prior = em_ptr->smooth ? em_ptr->compute_log_prior() : 0.0;
lambda = likelihood + log_prior;
if (verb_em) {
if (em_ptr->smooth) {
prism_printf("Iteration #%d:\tlog_likelihood=%.9f\tlog_prior=%.9f\tlog_post=%.9f\n", iterate, likelihood, log_prior, lambda);
}
else {
prism_printf("Iteration #%d:\tlog_likelihood=%.9f\n", iterate, likelihood);
}
}
if (debug_level) {
prism_printf("After I-step[%d]:\n", iterate);
prism_printf("likelihood = %.9f\n", likelihood);
print_egraph(debug_level, PRINT_EM);
}
if (!isfinite(lambda)) {
emit_internal_error("invalid log likelihood or log post: %s (at iteration #%d)",
isnan(lambda) ? "NaN" : "infinity", iterate);
RET_ERR(ierr_invalid_likelihood);
}
if (old_valid && old_lambda - lambda > prism_epsilon) {
emit_error("log likelihood or log post decreased [old: %.9f, new: %.9f] (at iteration #%d)",
old_lambda, lambda, iterate);
RET_ERR(err_invalid_likelihood);
}
if (itemp == 1.0 && likelihood > 0.0) {
emit_error("log likelihood greater than zero [value: %.9f] (at iteration #%d)",
likelihood, iterate);
RET_ERR(err_invalid_likelihood);
}
converged = (old_valid && lambda - old_lambda <= prism_epsilon);
if (converged || REACHED_MAX_ITERATE(iterate)) {
break;
}
old_lambda = lambda;
old_valid = 1;
RET_ON_ERR(em_ptr->compute_expectation());
if (debug_level) {
prism_printf("After O-step[%d]:\n", iterate);
print_egraph(debug_level, PRINT_EM);
}
SHOW_PROGRESS(iterate);
RET_ON_ERR(em_ptr->update_params());
iterate++;
}
/* [21 Aug 2007, by yuizumi]
* Note that 1.0 can be represented exactly in IEEE 754.
*/
if (itemp == 1.0) {
break;
}
itemp *= itemp_rate;
if (itemp >= 1.0) {
itemp = 1.0;
}
}
SHOW_PROGRESS_TAIL(converged, iterate, lambda);
if (r == 0 || lambda > em_ptr->lambda) {
em_ptr->lambda = lambda;
em_ptr->likelihood = likelihood;
em_ptr->iterate = iterate;
saved = (r < num_restart - 1);
if (saved) {
save_params();
}
}
}
if (saved) {
restore_params();
}
em_ptr->bic = compute_bic(em_ptr->likelihood);
em_ptr->cs = em_ptr->smooth ? compute_cs(em_ptr->likelihood) : 0.0;
return BP_TRUE;
}
int mpm_run_em(EM_ENG_PTR emptr) {
int r, iterate, old_valid, converged, saved=0;
double likelihood, log_prior;
double lambda, old_lambda=0.0;
config_em(emptr);
for (r = 0; r < num_restart; r++) {
SHOW_PROGRESS_HEAD("#em-iters", r);
initialize_params();
mpm_bcast_inside();
clear_sw_msg_send();
itemp = daem ? itemp_init : 1.0;
iterate = 0;
while (1) {
if (daem) {
SHOW_PROGRESS_TEMP(itemp);
}
old_valid = 0;
while (1) {
if (CTRLC_PRESSED) {
SHOW_PROGRESS_INTR();
RET_ERR(err_ctrl_c_pressed);
}
if (failure_observed) {
inside_failure = mp_sum_value(0.0);
}
log_prior = emptr->smooth ? emptr->compute_log_prior() : 0.0;
lambda = mp_sum_value(log_prior);
likelihood = lambda - log_prior;
mp_debug("local lambda = %.9f, lambda = %.9f", log_prior, lambda);
if (verb_em) {
if (emptr->smooth) {
prism_printf("Iteration #%d:\tlog_likelihood=%.9f\tlog_prior=%.9f\tlog_post=%.9f\n", iterate, likelihood, log_prior, lambda);
} else {
prism_printf("Iteration #%d:\tlog_likelihood=%.9f\n", iterate, likelihood);
}
}
if (!isfinite(lambda)) {
emit_internal_error("invalid log likelihood or log post: %s (at iterateion #%d)",
isnan(lambda) ? "NaN" : "infinity", iterate);
RET_ERR(ierr_invalid_likelihood);
}
if (old_valid && old_lambda - lambda > prism_epsilon) {
emit_error("log likelihood or log post decreased [old: %.9f, new: %.9f] (at iteration #%d)",
old_lambda, lambda, iterate);
RET_ERR(err_invalid_likelihood);
}
if (itemp == 1.0 && likelihood > 0.0) {
emit_error("log likelihood greater than zero [value: %.9f] (at iteration #%d)",
likelihood, iterate);
RET_ERR(err_invalid_likelihood);
}
converged = (old_valid && lambda - old_lambda <= prism_epsilon);
if (converged || REACHED_MAX_ITERATE(iterate)) {
break;
}
old_lambda = lambda;
old_valid = 1;
mpm_share_expectation();
SHOW_PROGRESS(iterate);
RET_ON_ERR(emptr->update_params());
iterate++;
}
if (itemp == 1.0) {
break;
}
itemp *= itemp_rate;
if (itemp >= 1.0) {
itemp = 1.0;
}
}
SHOW_PROGRESS_TAIL(converged, iterate, lambda);
if (r == 0 || lambda > emptr->lambda) {
emptr->lambda = lambda;
emptr->likelihood = likelihood;
emptr->iterate = iterate;
saved = (r < num_restart - 1);
if (saved) {
save_params();
}
}
}
if (saved) {
restore_params();
}
emptr->bic = compute_bic(emptr->likelihood);
emptr->cs = emptr->smooth ? compute_cs(emptr->likelihood) : 0.0;
return BP_TRUE;
}
