/** * permute_choices * * Call append_choices() for each BLOB_CHOICE in BLOB_CHOICE_LIST * with the given char_choice_index in char_choices. */ void Dict::permute_choices( const char *debug, const BLOB_CHOICE_LIST_VECTOR &char_choices, int char_choice_index, const CHAR_FRAGMENT_INFO *prev_char_frag_info, WERD_CHOICE *word, float certainties[], float *limit, WERD_CHOICE *best_choice, int *attempts_left, void *more_args) { if (debug) { tprintf("%s permute_choices: char_choice_index=%d" " limit=%g rating=%g, certainty=%g word=%s\n", debug, char_choice_index, *limit, word->rating(), word->certainty(), word->debug_string().string()); } if (char_choice_index < char_choices.length()) { BLOB_CHOICE_IT blob_choice_it; blob_choice_it.set_to_list(char_choices.get(char_choice_index)); for (blob_choice_it.mark_cycle_pt(); !blob_choice_it.cycled_list(); blob_choice_it.forward()) { (*attempts_left)--; append_choices(debug, char_choices, *(blob_choice_it.data()), char_choice_index, prev_char_frag_info, word, certainties, limit, best_choice, attempts_left, more_args); if (*attempts_left <= 0) { if (debug) tprintf("permute_choices(): attempts_left is 0\n"); break; } } } }
/** * print_char_choices_list */ void print_char_choices_list(const char *msg, const BLOB_CHOICE_LIST_VECTOR &char_choices, const UNICHARSET ¤t_unicharset, BOOL8 detailed) { if (*msg != '\0') tprintf("%s\n", msg); for (int x = 0; x < char_choices.length(); ++x) { BLOB_CHOICE_IT c_it; c_it.set_to_list(char_choices.get(x)); tprintf("char[%d]: %s\n", x, current_unicharset.debug_str( c_it.data()->unichar_id()).string()); if (detailed) print_ratings_list(" ", char_choices.get(x), current_unicharset); } }
void Wordrec::CopyCharChoices(const BLOB_CHOICE_LIST_VECTOR &from, BLOB_CHOICE_LIST_VECTOR *to) { to->delete_data_pointers(); to->clear(); for (int i = 0; i < from.size(); ++i) { BLOB_CHOICE_LIST *cc_list = new BLOB_CHOICE_LIST(); cc_list->deep_copy(from[i], &BLOB_CHOICE::deep_copy); to->push_back(cc_list); } }
/** * append_choices * * Checks to see whether or not the next choice is worth appending to * the word being generated. If so then keeps going deeper into the word. * * This function assumes that Dict::go_deeper_fxn_ is set. */ void Dict::append_choices( const char *debug, const BLOB_CHOICE_LIST_VECTOR &char_choices, const BLOB_CHOICE &blob_choice, int char_choice_index, const CHAR_FRAGMENT_INFO *prev_char_frag_info, WERD_CHOICE *word, float certainties[], float *limit, WERD_CHOICE *best_choice, int *attempts_left, void *more_args) { int word_ending = (char_choice_index == char_choices.length() - 1) ? true : false; // Deal with fragments. CHAR_FRAGMENT_INFO char_frag_info; if (!fragment_state_okay(blob_choice.unichar_id(), blob_choice.rating(), blob_choice.certainty(), prev_char_frag_info, debug, word_ending, &char_frag_info)) { return; // blob_choice must be an invalid fragment } // Search the next letter if this character is a fragment. if (char_frag_info.unichar_id == INVALID_UNICHAR_ID) { permute_choices(debug, char_choices, char_choice_index + 1, &char_frag_info, word, certainties, limit, best_choice, attempts_left, more_args); return; } // Add the next unichar. float old_rating = word->rating(); float old_certainty = word->certainty(); uint8_t old_permuter = word->permuter(); certainties[word->length()] = char_frag_info.certainty; word->append_unichar_id_space_allocated( char_frag_info.unichar_id, char_frag_info.num_fragments, char_frag_info.rating, char_frag_info.certainty); // Explore the next unichar. (this->*go_deeper_fxn_)(debug, char_choices, char_choice_index, &char_frag_info, word_ending, word, certainties, limit, best_choice, attempts_left, more_args); // Remove the unichar we added to explore other choices in it's place. word->remove_last_unichar_id(); word->set_rating(old_rating); word->set_certainty(old_certainty); word->set_permuter(old_permuter); }
/** * @name go_deeper_dawg_fxn * * If the choice being composed so far could be a dictionary word * keep exploring choices. */ void Dict::go_deeper_dawg_fxn( const char *debug, const BLOB_CHOICE_LIST_VECTOR &char_choices, int char_choice_index, const CHAR_FRAGMENT_INFO *prev_char_frag_info, bool word_ending, WERD_CHOICE *word, float certainties[], float *limit, WERD_CHOICE *best_choice, int *attempts_left, void *void_more_args) { DawgArgs *more_args = static_cast<DawgArgs *>(void_more_args); word_ending = (char_choice_index == char_choices.size()-1); int word_index = word->length() - 1; if (best_choice->rating() < *limit) return; // Look up char in DAWG // If the current unichar is an ngram first try calling // letter_is_okay() for each unigram it contains separately. UNICHAR_ID orig_uch_id = word->unichar_id(word_index); bool checked_unigrams = false; if (getUnicharset().get_isngram(orig_uch_id)) { if (dawg_debug_level) { tprintf("checking unigrams in an ngram %s\n", getUnicharset().debug_str(orig_uch_id).string()); } int num_unigrams = 0; word->remove_last_unichar_id(); GenericVector<UNICHAR_ID> encoding; const char *ngram_str = getUnicharset().id_to_unichar(orig_uch_id); // Since the string came out of the unicharset, failure is impossible. ASSERT_HOST(getUnicharset().encode_string(ngram_str, true, &encoding, nullptr, nullptr)); bool unigrams_ok = true; // Construct DawgArgs that reflect the current state. DawgPositionVector unigram_active_dawgs = *(more_args->active_dawgs); DawgPositionVector unigram_updated_dawgs; DawgArgs unigram_dawg_args(&unigram_active_dawgs, &unigram_updated_dawgs, more_args->permuter); // Check unigrams in the ngram with letter_is_okay(). for (int i = 0; unigrams_ok && i < encoding.size(); ++i) { UNICHAR_ID uch_id = encoding[i]; ASSERT_HOST(uch_id != INVALID_UNICHAR_ID); ++num_unigrams; word->append_unichar_id(uch_id, 1, 0.0, 0.0); unigrams_ok = (this->*letter_is_okay_)( &unigram_dawg_args, word->unichar_id(word_index+num_unigrams-1), word_ending && i == encoding.size() - 1); (*unigram_dawg_args.active_dawgs) = *(unigram_dawg_args.updated_dawgs); if (dawg_debug_level) { tprintf("unigram %s is %s\n", getUnicharset().debug_str(uch_id).string(), unigrams_ok ? "OK" : "not OK"); } } // Restore the word and copy the updated dawg state if needed. while (num_unigrams-- > 0) word->remove_last_unichar_id(); word->append_unichar_id_space_allocated(orig_uch_id, 1, 0.0, 0.0); if (unigrams_ok) { checked_unigrams = true; more_args->permuter = unigram_dawg_args.permuter; *(more_args->updated_dawgs) = *(unigram_dawg_args.updated_dawgs); } } // Check which dawgs from the dawgs_ vector contain the word // up to and including the current unichar. if (checked_unigrams || (this->*letter_is_okay_)( more_args, word->unichar_id(word_index), word_ending)) { // Add a new word choice if (word_ending) { if (dawg_debug_level) { tprintf("found word = %s\n", word->debug_string().string()); } if (strcmp(output_ambig_words_file.string(), "") != 0) { if (output_ambig_words_file_ == nullptr) { output_ambig_words_file_ = fopen(output_ambig_words_file.string(), "wb+"); if (output_ambig_words_file_ == nullptr) { tprintf("Failed to open output_ambig_words_file %s\n", output_ambig_words_file.string()); exit(1); } STRING word_str; word->string_and_lengths(&word_str, nullptr); word_str += " "; fprintf(output_ambig_words_file_, "%s", word_str.string()); } STRING word_str; word->string_and_lengths(&word_str, nullptr); word_str += " "; fprintf(output_ambig_words_file_, "%s", word_str.string()); } WERD_CHOICE *adjusted_word = word; adjusted_word->set_permuter(more_args->permuter); update_best_choice(*adjusted_word, best_choice); } else { // search the next letter // Make updated_* point to the next entries in the DawgPositionVector // arrays (that were originally created in dawg_permute_and_select) ++(more_args->updated_dawgs); // Make active_dawgs and constraints point to the updated ones. ++(more_args->active_dawgs); permute_choices(debug, char_choices, char_choice_index + 1, prev_char_frag_info, word, certainties, limit, best_choice, attempts_left, more_args); // Restore previous state to explore another letter in this position. --(more_args->updated_dawgs); --(more_args->active_dawgs); } } else { if (dawg_debug_level) { tprintf("last unichar not OK at index %d in %s\n", word_index, word->debug_string().string()); } } }
/********************************************************************** * select_blob_to_split * * These are the results of the last classification. Find a likely * place to apply splits. **********************************************************************/ inT16 Wordrec::select_blob_to_split(const BLOB_CHOICE_LIST_VECTOR &char_choices, float rating_ceiling, bool split_next_to_fragment) { BLOB_CHOICE_IT blob_choice_it; BLOB_CHOICE *blob_choice; BLOB_CHOICE_IT temp_it; int x; float worst = -MAX_FLOAT32; int worst_index = -1; float worst_near_fragment = -MAX_FLOAT32; int worst_index_near_fragment = -1; const CHAR_FRAGMENT **fragments = NULL; if (chop_debug) { if (rating_ceiling < MAX_FLOAT32) cprintf("rating_ceiling = %8.4f\n", rating_ceiling); else cprintf("rating_ceiling = No Limit\n"); } if (split_next_to_fragment && char_choices.length() > 0) { fragments = new const CHAR_FRAGMENT *[char_choices.length()]; if (char_choices.get(0) != NULL) { temp_it.set_to_list(char_choices.get(0)); fragments[0] = getDict().getUnicharset().get_fragment( temp_it.data()->unichar_id()); } else { fragments[0] = NULL; } } for (x = 0; x < char_choices.length(); ++x) { if (char_choices.get(x) == NULL) { if (fragments != NULL) { delete[] fragments; } return x; } else { blob_choice_it.set_to_list(char_choices.get(x)); blob_choice = blob_choice_it.data(); // Populate fragments for the following position. if (split_next_to_fragment && x+1 < char_choices.length()) { if (char_choices.get(x+1) != NULL) { temp_it.set_to_list(char_choices.get(x+1)); fragments[x+1] = getDict().getUnicharset().get_fragment( temp_it.data()->unichar_id()); } else { fragments[x+1] = NULL; } } if (blob_choice->rating() < rating_ceiling && blob_choice->certainty() < tessedit_certainty_threshold) { // Update worst and worst_index. if (blob_choice->rating() > worst) { worst_index = x; worst = blob_choice->rating(); } if (split_next_to_fragment) { // Update worst_near_fragment and worst_index_near_fragment. bool expand_following_fragment = (x + 1 < char_choices.length() && fragments[x+1] != NULL && !fragments[x+1]->is_beginning()); bool expand_preceding_fragment = (x > 0 && fragments[x-1] != NULL && !fragments[x-1]->is_ending()); if ((expand_following_fragment || expand_preceding_fragment) && blob_choice->rating() > worst_near_fragment) { worst_index_near_fragment = x; worst_near_fragment = blob_choice->rating(); if (chop_debug) { cprintf("worst_index_near_fragment=%d" " expand_following_fragment=%d" " expand_preceding_fragment=%d\n", worst_index_near_fragment, expand_following_fragment, expand_preceding_fragment); } } } } } } if (fragments != NULL) { delete[] fragments; } // TODO(daria): maybe a threshold of badness for // worst_near_fragment would be useful. return worst_index_near_fragment != -1 ? worst_index_near_fragment : worst_index; }