1
0
forked from GitHub/gf-core

patch for adjustable heuristics from Python

This commit is contained in:
kr.angelov
2013-06-26 07:36:03 +00:00
parent d94b6146f2
commit 3c2d1890d0
9 changed files with 48 additions and 183 deletions

View File

@@ -121,8 +121,7 @@ libpgf_la_SOURCES = \
bin_PROGRAMS = \
utils/pgf-print \
utils/pgf-translate \
utils/pgf-parse \
utils/pgf-chunk
utils/pgf-parse
utils_pgf_print_SOURCES = utils/pgf-print.c
utils_pgf_print_LDADD = libpgf.la libgu.la
@@ -133,9 +132,6 @@ utils_pgf_translate_LDADD = libpgf.la libgu.la
utils_pgf_parse_SOURCES = utils/pgf-parse.c
utils_pgf_parse_LDADD = libpgf.la libgu.la
utils_pgf_chunk_SOURCES = utils/pgf-chunk.c
utils_pgf_chunk_LDADD = libpgf.la libgu.la
AUTOMAKE_OPTIONS = foreign subdir-objects dist-bzip2
ACLOCAL_AMFLAGS = -I m4
include doxygen.am

View File

@@ -1638,7 +1638,8 @@ pgf_parsing_default_beam_size(PgfConcr* concr)
}
static PgfParsing*
pgf_new_parsing(PgfConcr* concr, GuPool* pool, GuPool* out_pool)
pgf_new_parsing(PgfConcr* concr, double heuristics,
GuPool* pool, GuPool* out_pool)
{
PgfParsing* ps = gu_new(PgfParsing, pool);
ps->concr = concr;
@@ -1654,7 +1655,7 @@ pgf_new_parsing(PgfConcr* concr, GuPool* pool, GuPool* out_pool)
ps->prod_full_count = 0;
#endif
ps->free_item = NULL;
ps->beam_size = pgf_parsing_default_beam_size(concr);
ps->beam_size = heuristics;
PgfExprMeta *expr_meta =
gu_new_variant(PGF_EXPR_META,
@@ -2214,6 +2215,7 @@ pgf_parse_print_chunks(PgfParseState* state)
// TODO: s/CId/Cat, add the cid to Cat, make Cat the key to CncCat
PgfParseState*
pgf_parser_init_state(PgfConcr* concr, PgfCId cat, size_t lin_idx,
double heuristics,
GuPool* pool, GuPool* out_pool)
{
PgfCncCat* cnccat =
@@ -2223,8 +2225,12 @@ pgf_parser_init_state(PgfConcr* concr, PgfCId cat, size_t lin_idx,
gu_assert(lin_idx < cnccat->n_lins);
if (heuristics < 0) {
heuristics = pgf_parsing_default_beam_size(concr);
}
PgfParsing* ps =
pgf_new_parsing(concr, pool, out_pool);
pgf_new_parsing(concr, heuristics, pool, out_pool);
PgfParseState* state =
pgf_new_parse_state(ps, NULL, NULL, pool);
@@ -2269,12 +2275,6 @@ pgf_parser_init_state(PgfConcr* concr, PgfCId cat, size_t lin_idx,
return state;
}
void
pgf_parser_set_beam_size(PgfParseState* state, double beam_size)
{
state->ps->beam_size = beam_size;
}
void
pgf_parser_add_literal(PgfConcr *concr, PgfCId cat,
PgfLiteralCallback* callback)

View File

@@ -34,6 +34,7 @@ typedef struct PgfParseState PgfParseState;
/// Begin parsing
PgfParseState*
pgf_parser_init_state(PgfConcr* concr, PgfCId cat, size_t lin_idx,
double heuristics,
GuPool* pool, GuPool* out_pool);
/**<
* @param parser The parser to use
@@ -69,9 +70,6 @@ pgf_parser_next_state(PgfParseState* prev, PgfToken tok);
GuEnum*
pgf_parser_completions(PgfParseState* prev, GuString prefix);
void
pgf_parser_set_beam_size(PgfParseState* state, double beam_size);
void
pgf_parser_add_literal(PgfConcr *concr, PgfCId cat,
PgfLiteralCallback* callback);

View File

@@ -157,7 +157,7 @@ pgf_parseval(PgfConcr* concr, PgfExpr expr, PgfCId cat,
PgfMetricsLznState state;
state.funcs = &pgf_metrics_lin_funcs1;
state.ps = pgf_parser_init_state(concr, cat, 0, pool, pool);
state.ps = pgf_parser_init_state(concr, cat, 0, -1, pool, pool);
state.marks = gu_new_buf(int, pool);
state.pos = 0;
state.phrases = gu_new_buf(PgfPhrase*, pool);

View File

@@ -210,10 +210,18 @@ pgf_linearize(PgfConcr* concr, PgfExpr expr, GuWriter* wtr, GuExn* err)
GuEnum*
pgf_parse(PgfConcr* concr, PgfCId cat, PgfLexer *lexer,
GuPool* pool, GuPool* out_pool)
{
return pgf_parse_with_heuristics(concr, cat, lexer, -1.0, pool, out_pool);
}
GuEnum*
pgf_parse_with_heuristics(PgfConcr* concr, PgfCId cat, PgfLexer *lexer,
double heuristics,
GuPool* pool, GuPool* out_pool)
{
// Begin parsing a sentence of the specified category
PgfParseState* state =
pgf_parser_init_state(concr, cat, 0, pool, out_pool);
pgf_parser_init_state(concr, cat, 0, heuristics, pool, out_pool);
if (state == NULL) {
return NULL;
}
@@ -244,7 +252,7 @@ pgf_get_completions(PgfConcr* concr, PgfCId cat, PgfLexer *lexer,
{
// Begin parsing a sentence of the specified category
PgfParseState* state =
pgf_parser_init_state(concr, cat, 0, pool, pool);
pgf_parser_init_state(concr, cat, 0, -1, pool, pool);
if (state == NULL) {
return NULL;
}
@@ -268,31 +276,3 @@ pgf_get_completions(PgfConcr* concr, PgfCId cat, PgfLexer *lexer,
// Now begin enumerating the resulting syntax trees
return pgf_parser_completions(state, prefix);
}
void
pgf_print_chunks(PgfConcr* concr, PgfCId cat, PgfLexer *lexer, GuPool* pool)
{
// Begin parsing a sentence of the specified category
PgfParseState* state =
pgf_parser_init_state(concr, cat, 0, pool, pool);
if (state == NULL) {
printf("\n");
return;
}
// Tokenization
GuExn* lex_err = gu_new_exn(NULL, gu_kind(type), pool);
PgfToken tok = pgf_lexer_read_token(lexer, lex_err);
while (!gu_exn_is_raised(lex_err)) {
// feed the token to get a new parse state
state = pgf_parser_next_state(state, tok);
if (state == NULL) {
printf("\n");
return;
}
tok = pgf_lexer_read_token(lexer, lex_err);
}
pgf_parse_print_chunks(state);
}

View File

@@ -117,6 +117,11 @@ PgfExprEnum*
pgf_parse(PgfConcr* concr, PgfCId cat, PgfLexer *lexer,
GuPool* pool, GuPool* out_pool);
PgfExprEnum*
pgf_parse_with_heuristics(PgfConcr* concr, PgfCId cat, PgfLexer *lexer,
double heuristics,
GuPool* pool, GuPool* out_pool);
GuEnum*
pgf_get_completions(PgfConcr* concr, PgfCId cat, PgfLexer *lexer,
GuString prefix, GuPool* pool);
@@ -128,11 +133,6 @@ pgf_parseval(PgfConcr* concr, PgfExpr expr, PgfCId cat,
PgfExprEnum*
pgf_generate(PgfPGF* pgf, PgfCId cat, GuPool* pool);
// an experimental function. Please don't use it
void
pgf_print_chunks(PgfConcr* concr, PgfCId cat, PgfLexer *lexer, GuPool* pool);
/// @}
void

View File

@@ -1,112 +0,0 @@
// Don't give too much hope to this script. It is doing the wrong thing
// but let's see how far we can get with it.
#include <gu/variant.h>
#include <gu/map.h>
#include <gu/dump.h>
#include <gu/log.h>
#include <gu/enum.h>
#include <gu/file.h>
#include <pgf/pgf.h>
#include <pgf/parser.h>
#include <pgf/lexer.h>
#include <pgf/literals.h>
#include <pgf/linearizer.h>
#include <pgf/expr.h>
#include <pgf/edsl.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <locale.h>
#include <time.h>
int main(int argc, char* argv[]) {
// Set the character locale, so we can produce proper output.
setlocale(LC_CTYPE, "");
// Create the pool that is used to allocate everything
GuPool* pool = gu_new_pool();
int status = EXIT_SUCCESS;
if (argc != 4) {
fprintf(stderr, "usage: %s pgf cat from_lang\n", argv[0]);
status = EXIT_FAILURE;
goto fail;
}
char* filename = argv[1];
GuString cat = gu_str_string(argv[2], pool);
GuString from_lang = gu_str_string(argv[3], pool);
// Create an exception frame that catches all errors.
GuExn* err = gu_new_exn(NULL, gu_kind(type), pool);
// Read the PGF grammar.
PgfPGF* pgf = pgf_read(filename, pool, err);
// If an error occured, it shows in the exception frame
if (!gu_ok(err)) {
fprintf(stderr, "Reading PGF failed\n");
status = EXIT_FAILURE;
goto fail;
}
pgf_load_meta_child_probs(pgf, "../../../treebanks/PennTreebank/ParseEngAbs3.probs", pool, err);
if (!gu_ok(err)) {
fprintf(stderr, "Loading meta child probs failed\n");
status = EXIT_FAILURE;
goto fail;
}
// Look up the source and destination concrete categories
PgfConcr* from_concr = pgf_get_language(pgf, from_lang);
if (!from_concr) {
fprintf(stderr, "Unknown language\n");
status = EXIT_FAILURE;
goto fail_concr;
}
// Register a callback for the literal category Symbol
pgf_parser_add_literal(from_concr, gu_str_string("Symb", pool),
&pgf_nerc_literal_callback);
// We will keep the latest results in the 'ppool' and
// we will iterate over them by using 'result'.
GuPool* ppool = NULL;
// The interactive translation loop.
// XXX: This currently reads stdin directly, so it doesn't support
// encodings properly. TODO: use a locale reader for input
while (true) {
char buf[4096];
char* line = fgets(buf, sizeof(buf), stdin);
if (line == NULL) {
if (ferror(stdin)) {
fprintf(stderr, "Input error\n");
status = EXIT_FAILURE;
}
break;
} else if (strcmp(line, "") == 0) {
// End nicely on empty input
break;
}
// We create a temporary pool for translating a single
// sentence, so our memory usage doesn't increase over time.
ppool = gu_new_pool();
GuReader *rdr =
gu_string_reader(gu_str_string(line, ppool), ppool);
PgfLexer *lexer =
pgf_new_simple_lexer(rdr, ppool);
pgf_print_chunks(from_concr, cat, lexer, ppool);
// Free all resources allocated during parsing and linearization
gu_pool_free(ppool);
}
fail_concr:
fail:
gu_pool_free(pool);
return status;
}

View File

@@ -25,8 +25,8 @@ int main(int argc, char* argv[]) {
// Create the pool that is used to allocate everything
GuPool* pool = gu_new_pool();
int status = EXIT_SUCCESS;
if (argc != 4) {
fprintf(stderr, "usage: %s pgf-file start-cat cnc-lang\n", argv[0]);
if (argc < 4 || argc > 5) {
fprintf(stderr, "usage: %s pgf-file start-cat cnc-lang [heuristics]\n(0.0 <= heuristics < 1.0, default: 0.95)\n", argv[0]);
status = EXIT_FAILURE;
goto fail;
}
@@ -34,6 +34,11 @@ int main(int argc, char* argv[]) {
GuString cat = gu_str_string(argv[2], pool);
GuString lang = gu_str_string(argv[3], pool);
double heuristics = 0.95;
if (argc == 5) {
heuristics = atof(argv[4]);
}
// Create an exception frame that catches all errors.
GuExn* err = gu_new_exn(NULL, gu_kind(type), pool);
@@ -65,7 +70,7 @@ int main(int argc, char* argv[]) {
clock_t end = clock();
double cpu_time_used = ((double) (end - start)) / CLOCKS_PER_SEC;
fprintf(stderr, "(%.0f ms) Ready to parse!\n", 1000.0 * cpu_time_used);
fprintf(stderr, "(%.0f ms) Ready to parse [heuristics=%.2f]!\n", 1000.0 * cpu_time_used, heuristics);
// Create an output stream for stdout
GuOut* out = gu_file_out(stdout, pool);
@@ -113,18 +118,9 @@ int main(int argc, char* argv[]) {
clock_t start = clock();
// Begin parsing a sentence of the specified category
PgfParseState* state =
pgf_parser_init_state(concr, cat, 0, ppool, ppool);
if (state == NULL) {
fprintf(stderr, "Couldn't begin parsing\n");
status = EXIT_FAILURE;
break;
}
GuReader *rdr = gu_string_reader(gu_str_string(line, ppool), ppool);
PgfLexer *lexer = pgf_new_simple_lexer(rdr, ppool);
GuEnum* result = pgf_parse(concr, cat, lexer, ppool, ppool);
GuEnum* result = pgf_parse_with_heuristics(concr, cat, lexer, heuristics, ppool, ppool);
PgfExprProb* ep = NULL;
if (result != NULL)

View File

@@ -692,15 +692,16 @@ void pypgf_container_descructor(PyObject *capsule)
static IterObject*
Concr_parse(ConcrObject* self, PyObject *args, PyObject *keywds)
{
static char *kwlist[] = {"sentence", "tokens", "cat", "n", NULL};
static char *kwlist[] = {"sentence", "tokens", "cat", "n", "heuristics", NULL};
int len;
const uint8_t *buf = NULL;
PyObject* py_lexer = NULL;
const char *catname_s = NULL;
int max_count = -1;
if (!PyArg_ParseTupleAndKeywords(args, keywds, "|s#Osi", kwlist,
&buf, &len, &py_lexer, &catname_s, &max_count))
double heuristics = -1;
if (!PyArg_ParseTupleAndKeywords(args, keywds, "|s#Osid", kwlist,
&buf, &len, &py_lexer, &catname_s, &max_count, &heuristics))
return NULL;
if ((buf == NULL && py_lexer == NULL) ||
@@ -752,7 +753,8 @@ Concr_parse(ConcrObject* self, PyObject *args, PyObject *keywds)
}
pyres->res =
pgf_parse(self->concr, catname, lexer, pyres->pool, out_pool);
pgf_parse_with_heuristics(self->concr, catname, lexer,
heuristics, pyres->pool, out_pool);
if (pyres->res == NULL) {
Py_DECREF(pyres);
@@ -1217,7 +1219,12 @@ static PyMethodDef Concr_methods[] = {
"Returns the print name of a function or category"
},
{"parse", (PyCFunction)Concr_parse, METH_VARARGS | METH_KEYWORDS,
"Parses a string and returns an iterator over the abstract trees for this sentence"
"Parses a string and returns an iterator over the abstract trees for this sentence\n\n"
"Named arguments:\n"
"- sentence (string) or tokens (list of strings)\n"
"- cat (string); OPTIONAL, default: the startcat of the grammar\n"
"- n (int), max. trees; OPTIONAL, default: extract all trees\n"
"- heuristics (double >= 0.0); OPTIONAL, default: taken from the flags in the grammar"
},
{"getCompletions", (PyCFunction)Concr_getCompletions, METH_VARARGS | METH_KEYWORDS,
"Parses a partial string and returns a list with the top n possible next tokens"