343 lines
8.7 KiB
C++
343 lines
8.7 KiB
C++
/*
|
|
* Copyright (C) 2009 The Android Open Source Project
|
|
*
|
|
* Licensed under the Apache License, Version 2.0 (the "License");
|
|
* you may not use this file except in compliance with the License.
|
|
* You may obtain a copy of the License at
|
|
*
|
|
* http://www.apache.org/licenses/LICENSE-2.0
|
|
*
|
|
* Unless required by applicable law or agreed to in writing, software
|
|
* distributed under the License is distributed on an "AS IS" BASIS,
|
|
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
* See the License for the specific language governing permissions and
|
|
* limitations under the License.
|
|
*/
|
|
|
|
#include <assert.h>
|
|
#include <math.h>
|
|
#include <stdio.h>
|
|
#include <string.h>
|
|
#include <time.h>
|
|
#include "mystdlib.h"
|
|
#include "ngram.h"
|
|
|
|
namespace ime_pinyin {
|
|
|
|
#define ADD_COUNT 0.3
|
|
|
|
int comp_double(const void *p1, const void *p2) {
|
|
if (*static_cast<const double*>(p1) < *static_cast<const double*>(p2))
|
|
return -1;
|
|
if (*static_cast<const double*>(p1) > *static_cast<const double*>(p2))
|
|
return 1;
|
|
return 0;
|
|
}
|
|
|
|
inline double distance(double freq, double code) {
|
|
// return fabs(freq - code);
|
|
return freq * fabs(log(freq) - log(code));
|
|
}
|
|
|
|
// Find the index of the code value which is nearest to the given freq
|
|
int qsearch_nearest(double code_book[], double freq, int start, int end) {
|
|
if (start == end)
|
|
return start;
|
|
|
|
if (start + 1 == end) {
|
|
if (distance(freq, code_book[end]) > distance(freq, code_book[start]))
|
|
return start;
|
|
return end;
|
|
}
|
|
|
|
int mid = (start + end) / 2;
|
|
|
|
if (code_book[mid] > freq)
|
|
return qsearch_nearest(code_book, freq, start, mid);
|
|
else
|
|
return qsearch_nearest(code_book, freq, mid, end);
|
|
}
|
|
|
|
size_t update_code_idx(double freqs[], size_t num, double code_book[],
|
|
CODEBOOK_TYPE *code_idx) {
|
|
size_t changed = 0;
|
|
for (size_t pos = 0; pos < num; pos++) {
|
|
CODEBOOK_TYPE idx;
|
|
idx = qsearch_nearest(code_book, freqs[pos], 0, kCodeBookSize - 1);
|
|
if (idx != code_idx[pos])
|
|
changed++;
|
|
code_idx[pos] = idx;
|
|
}
|
|
return changed;
|
|
}
|
|
|
|
double recalculate_kernel(double freqs[], size_t num, double code_book[],
|
|
CODEBOOK_TYPE *code_idx) {
|
|
double ret = 0;
|
|
|
|
size_t *item_num = new size_t[kCodeBookSize];
|
|
assert(item_num);
|
|
memset(item_num, 0, sizeof(size_t) * kCodeBookSize);
|
|
|
|
double *cb_new = new double[kCodeBookSize];
|
|
assert(cb_new);
|
|
memset(cb_new, 0, sizeof(double) * kCodeBookSize);
|
|
|
|
for (size_t pos = 0; pos < num; pos++) {
|
|
ret += distance(freqs[pos], code_book[code_idx[pos]]);
|
|
|
|
cb_new[code_idx[pos]] += freqs[pos];
|
|
item_num[code_idx[pos]] += 1;
|
|
}
|
|
|
|
for (size_t code = 0; code < kCodeBookSize; code++) {
|
|
assert(item_num[code] > 0);
|
|
code_book[code] = cb_new[code] / item_num[code];
|
|
}
|
|
|
|
delete [] item_num;
|
|
delete [] cb_new;
|
|
|
|
return ret;
|
|
}
|
|
|
|
void iterate_codes(double freqs[], size_t num, double code_book[],
|
|
CODEBOOK_TYPE *code_idx) {
|
|
size_t iter_num = 0;
|
|
double delta_last = 0;
|
|
do {
|
|
size_t changed = update_code_idx(freqs, num, code_book, code_idx);
|
|
|
|
double delta = recalculate_kernel(freqs, num, code_book, code_idx);
|
|
|
|
if (kPrintDebug0) {
|
|
printf("---Unigram codebook iteration: %d : %d, %.9f\n",
|
|
iter_num, changed, delta);
|
|
}
|
|
iter_num++;
|
|
|
|
if (iter_num > 1 &&
|
|
(delta == 0 || fabs(delta_last - delta)/fabs(delta) < 0.000000001))
|
|
break;
|
|
delta_last = delta;
|
|
} while (true);
|
|
}
|
|
|
|
|
|
NGram* NGram::instance_ = NULL;
|
|
|
|
NGram::NGram() {
|
|
initialized_ = false;
|
|
idx_num_ = 0;
|
|
lma_freq_idx_ = NULL;
|
|
sys_score_compensation_ = 0;
|
|
|
|
#ifdef ___BUILD_MODEL___
|
|
freq_codes_df_ = NULL;
|
|
#endif
|
|
freq_codes_ = NULL;
|
|
}
|
|
|
|
NGram::~NGram() {
|
|
if (NULL != lma_freq_idx_)
|
|
free(lma_freq_idx_);
|
|
|
|
#ifdef ___BUILD_MODEL___
|
|
if (NULL != freq_codes_df_)
|
|
free(freq_codes_df_);
|
|
#endif
|
|
|
|
if (NULL != freq_codes_)
|
|
free(freq_codes_);
|
|
}
|
|
|
|
NGram& NGram::get_instance() {
|
|
if (NULL == instance_)
|
|
instance_ = new NGram();
|
|
return *instance_;
|
|
}
|
|
|
|
bool NGram::save_ngram(FILE *fp) {
|
|
if (!initialized_ || NULL == fp)
|
|
return false;
|
|
|
|
if (0 == idx_num_ || NULL == freq_codes_ || NULL == lma_freq_idx_)
|
|
return false;
|
|
|
|
if (fwrite(&idx_num_, sizeof(uint32), 1, fp) != 1)
|
|
return false;
|
|
|
|
if (fwrite(freq_codes_, sizeof(LmaScoreType), kCodeBookSize, fp) !=
|
|
kCodeBookSize)
|
|
return false;
|
|
|
|
if (fwrite(lma_freq_idx_, sizeof(CODEBOOK_TYPE), idx_num_, fp) != idx_num_)
|
|
return false;
|
|
|
|
return true;
|
|
}
|
|
|
|
bool NGram::load_ngram(FILE *fp) {
|
|
if (NULL == fp)
|
|
return false;
|
|
|
|
initialized_ = false;
|
|
|
|
if (fread(&idx_num_, sizeof(uint32), 1, fp) != 1 )
|
|
return false;
|
|
|
|
if (NULL != lma_freq_idx_)
|
|
free(lma_freq_idx_);
|
|
|
|
if (NULL != freq_codes_)
|
|
free(freq_codes_);
|
|
|
|
lma_freq_idx_ = static_cast<CODEBOOK_TYPE*>
|
|
(malloc(idx_num_ * sizeof(CODEBOOK_TYPE)));
|
|
freq_codes_ = static_cast<LmaScoreType*>
|
|
(malloc(kCodeBookSize * sizeof(LmaScoreType)));
|
|
|
|
if (NULL == lma_freq_idx_ || NULL == freq_codes_)
|
|
return false;
|
|
|
|
if (fread(freq_codes_, sizeof(LmaScoreType), kCodeBookSize, fp) !=
|
|
kCodeBookSize)
|
|
return false;
|
|
|
|
if (fread(lma_freq_idx_, sizeof(CODEBOOK_TYPE), idx_num_, fp) != idx_num_)
|
|
return false;
|
|
|
|
initialized_ = true;
|
|
|
|
total_freq_none_sys_ = 0;
|
|
return true;
|
|
}
|
|
|
|
void NGram::set_total_freq_none_sys(size_t freq_none_sys) {
|
|
total_freq_none_sys_ = freq_none_sys;
|
|
if (0 == total_freq_none_sys_) {
|
|
sys_score_compensation_ = 0;
|
|
} else {
|
|
double factor = static_cast<double>(kSysDictTotalFreq) / (
|
|
kSysDictTotalFreq + total_freq_none_sys_);
|
|
sys_score_compensation_ = static_cast<float>(
|
|
log(factor) * kLogValueAmplifier);
|
|
}
|
|
}
|
|
|
|
// The caller makes sure this oject is initialized.
|
|
float NGram::get_uni_psb(LemmaIdType lma_id) {
|
|
return static_cast<float>(freq_codes_[lma_freq_idx_[lma_id]]) +
|
|
sys_score_compensation_;
|
|
}
|
|
|
|
float NGram::convert_psb_to_score(double psb) {
|
|
float score = static_cast<float>(
|
|
log(psb) * static_cast<double>(kLogValueAmplifier));
|
|
if (score > static_cast<float>(kMaxScore)) {
|
|
score = static_cast<float>(kMaxScore);
|
|
}
|
|
return score;
|
|
}
|
|
|
|
#ifdef ___BUILD_MODEL___
|
|
bool NGram::build_unigram(LemmaEntry *lemma_arr, size_t lemma_num,
|
|
LemmaIdType next_idx_unused) {
|
|
if (NULL == lemma_arr || 0 == lemma_num || next_idx_unused <= 1)
|
|
return false;
|
|
|
|
double total_freq = 0;
|
|
double *freqs = new double[next_idx_unused];
|
|
if (NULL == freqs)
|
|
return false;
|
|
|
|
freqs[0] = ADD_COUNT;
|
|
total_freq += freqs[0];
|
|
LemmaIdType idx_now = 0;
|
|
for (size_t pos = 0; pos < lemma_num; pos++) {
|
|
if (lemma_arr[pos].idx_by_hz == idx_now)
|
|
continue;
|
|
idx_now++;
|
|
|
|
assert(lemma_arr[pos].idx_by_hz == idx_now);
|
|
|
|
freqs[idx_now] = lemma_arr[pos].freq;
|
|
if (freqs[idx_now] <= 0)
|
|
freqs[idx_now] = 0.3;
|
|
|
|
total_freq += freqs[idx_now];
|
|
}
|
|
|
|
double max_freq = 0;
|
|
idx_num_ = idx_now + 1;
|
|
assert(idx_now + 1 == next_idx_unused);
|
|
|
|
for (size_t pos = 0; pos < idx_num_; pos++) {
|
|
freqs[pos] = freqs[pos] / total_freq;
|
|
assert(freqs[pos] > 0);
|
|
if (freqs[pos] > max_freq)
|
|
max_freq = freqs[pos];
|
|
}
|
|
|
|
// calculate the code book
|
|
if (NULL == freq_codes_df_)
|
|
freq_codes_df_ = new double[kCodeBookSize];
|
|
assert(freq_codes_df_);
|
|
memset(freq_codes_df_, 0, sizeof(double) * kCodeBookSize);
|
|
|
|
if (NULL == freq_codes_)
|
|
freq_codes_ = new LmaScoreType[kCodeBookSize];
|
|
assert(freq_codes_);
|
|
memset(freq_codes_, 0, sizeof(LmaScoreType) * kCodeBookSize);
|
|
|
|
size_t freq_pos = 0;
|
|
for (size_t code_pos = 0; code_pos < kCodeBookSize; code_pos++) {
|
|
bool found = true;
|
|
|
|
while (found) {
|
|
found = false;
|
|
double cand = freqs[freq_pos];
|
|
for (size_t i = 0; i < code_pos; i++)
|
|
if (freq_codes_df_[i] == cand) {
|
|
found = true;
|
|
break;
|
|
}
|
|
if (found)
|
|
freq_pos++;
|
|
}
|
|
|
|
freq_codes_df_[code_pos] = freqs[freq_pos];
|
|
freq_pos++;
|
|
}
|
|
|
|
myqsort(freq_codes_df_, kCodeBookSize, sizeof(double), comp_double);
|
|
|
|
if (NULL == lma_freq_idx_)
|
|
lma_freq_idx_ = new CODEBOOK_TYPE[idx_num_];
|
|
assert(lma_freq_idx_);
|
|
|
|
iterate_codes(freqs, idx_num_, freq_codes_df_, lma_freq_idx_);
|
|
|
|
delete [] freqs;
|
|
|
|
if (kPrintDebug0) {
|
|
printf("\n------Language Model Unigram Codebook------\n");
|
|
}
|
|
|
|
for (size_t code_pos = 0; code_pos < kCodeBookSize; code_pos++) {
|
|
double log_score = log(freq_codes_df_[code_pos]);
|
|
float final_score = convert_psb_to_score(freq_codes_df_[code_pos]);
|
|
if (kPrintDebug0) {
|
|
printf("code:%d, probability:%.9f, log score:%.3f, final score: %.3f\n",
|
|
code_pos, freq_codes_df_[code_pos], log_score, final_score);
|
|
}
|
|
freq_codes_[code_pos] = static_cast<LmaScoreType>(final_score);
|
|
}
|
|
|
|
initialized_ = true;
|
|
return true;
|
|
}
|
|
#endif
|
|
|
|
} // namespace ime_pinyin
|