void calc_mem_req(ng_t *ng,flag is_ascii) {
ngram current_ngram;
ngram previous_ngram;
count_t *ng_count;
int i,j;
current_ngram.id_array = (id__t *) rr_malloc(sizeof(id__t)*ng->n);
previous_ngram.id_array = (id__t *) rr_malloc(sizeof(id__t)*ng->n);
ng_count = (count_t *) rr_calloc(ng->n,sizeof(count_t));
current_ngram.n = ng->n;
rewind(ng->id_gram_fp);
while (!rr_feof(ng->id_gram_fp)) {
for (i=0;i<=ng->n-1;i++) {
previous_ngram.id_array[i]=current_ngram.id_array[i];
}
get_ngram(ng->id_gram_fp,¤t_ngram,is_ascii);
for (i=0;i<=ng->n-1;i++) {
if (current_ngram.id_array[i] != previous_ngram.id_array[i]) {
for (j=i;j<=ng->n-1;j++) {
if (j>0) {
if (ng_count[j] > ng->cutoffs[j-1]) {
ng->table_sizes[j]++;
}
}
ng_count[j] = current_ngram.count;
}
i=ng->n;
}
else {
ng_count[i] += current_ngram.count;
}
}
}
for (i=1;i<=ng->n-1;i++) {
if (ng_count[i] > ng->cutoffs[i-1]) {
ng->table_sizes[i]++;
}
}
/* Add a fudge factor, as problems can crop up with having to
cut-off last few n-grams. */
for (i=1;i<=ng->n-1;i++) {
ng->table_sizes[i]+=10;
}
rr_iclose(ng->id_gram_fp);
ng->id_gram_fp = rr_iopen(ng->id_gram_filename);
}
void compute_perplexity(ng_t *ng,
arpa_lm_t *arpa_ng,
char *text_stream_filename,
char *probs_stream_filename,
char *annotation_filename,
char *oov_filename,
char *fb_list_filename,
flag backoff_from_unk_inc,
flag backoff_from_unk_exc,
flag backoff_from_ccs_inc,
flag backoff_from_ccs_exc,
flag arpa_lm,
flag include_unks,
double log_base) {
fb_info *fb_list;
FILE *temp_fp;
FILE *text_stream_fp;
FILE *probs_stream_fp;
FILE *annotation_fp;
FILE *oov_fp;
flag out_probs;
flag annotate;
flag out_oovs;
flag found_unk_wrongly;
double prob;
double sum_log_prob;
int total_words;
int excluded_unks;
int excluded_ccs;
char current_word[1000]; /* Hope that's big enough */
char **prev_words;
vocab_sz_t current_id;
id__t short_current_id;
id__t *context;
int context_length;
int i;
int bo_case;
int actual_context_length;
int *ngrams_hit;
int n;
/* Initialise file pointers to prevent warnings from the compiler. */
probs_stream_fp = NULL;
annotation_fp = NULL;
oov_fp = NULL;
short_current_id = 0;
found_unk_wrongly = 0;
annotate = 0;
bo_case = 0;
if (arpa_lm) {
n = arpa_ng->n;
fb_list = gen_fb_list(arpa_ng->vocab_ht,
(int) arpa_ng->vocab_size,
arpa_ng->vocab,
arpa_ng->context_cue,
backoff_from_unk_inc,
backoff_from_unk_exc,
backoff_from_ccs_inc,
backoff_from_ccs_exc,
fb_list_filename);
}else {
n = ng->n;
fb_list = gen_fb_list(ng->vocab_ht,
(int) ng->vocab_size,
ng->vocab,
ng->context_cue,
backoff_from_unk_inc,
backoff_from_unk_exc,
backoff_from_ccs_inc,
backoff_from_ccs_exc,
fb_list_filename);
}
ngrams_hit = (int *) rr_calloc(n,sizeof(int));
prev_words = (char **) rr_malloc(sizeof(char *)*n);
for (i=0;i<=n-1;i++)
prev_words[i] = (char *) rr_malloc(sizeof(char)*1000);
/* Check that text_stream_filename and probs_stream_filename (if
specified) are valid. Note that the checks employed by the
standard rr_fopen tools are not suitable here, since we don't
want the program to terminate if the paths are not found. */
if (!strcmp(text_stream_filename,"")) {
printf("Error : Must specify a text file. Use the -text switch.\n");
return;
}
if (!rr_fexists(text_stream_filename) && strcmp(text_stream_filename,"-")) {
printf("Error : Can't open file %s for reading.\n",text_stream_filename);
return;
}
out_probs = strcmp(probs_stream_filename,"");
annotate = strcmp(annotation_filename,"");
out_oovs = strcmp(oov_filename,"");
printf("Computing perplexity of the language model with respect\n");
printf(" to the text %s\n",text_stream_filename);
if (out_probs)
printf("Probability stream will be written to file %s\n",
probs_stream_filename);
if (annotate)
printf("Annotation will be written to file %s\n",
annotation_filename);
if (out_oovs)
printf("Out of vocabulary words will be written to file %s\n",
oov_filename);
if (backoff_from_unk_inc)
printf("Will force inclusive back-off from OOVs.\n");
if (backoff_from_unk_exc)
printf("Will force exclusive back-off from OOVs.\n");
if (backoff_from_ccs_inc)
printf("Will force inclusive back-off from context cues.\n");
if (backoff_from_ccs_exc)
printf("Will force exclusive back-off from context cues.\n");
if (strcmp(fb_list_filename,""))
printf("Will force back-off according to the contents of %s\n",
fb_list_filename);
if (include_unks)
printf("Perplexity calculation will include OOVs.\n");
/* Check for existance of files, as rr functions will quit, which isn't
what we want */
if (out_probs && strcmp(probs_stream_filename,"-")) {
if ((temp_fp = fopen(probs_stream_filename,"w")) == NULL) {
printf("Error : Can't open file %s for writing.\n",probs_stream_filename);
return;
}
fclose(temp_fp);
}
if (annotate && strcmp(annotation_filename,"-")) {
if ((temp_fp = fopen(annotation_filename,"w")) == NULL) {
printf("Error : Can't open file %s for writing.\n",annotation_filename);
return;
}
fclose(temp_fp);
}
if (out_oovs && strcmp(oov_filename,"-")) {
if ((temp_fp = fopen(oov_filename,"w")) == NULL) {
printf("Error : Can't open file %s for writing.\n",oov_filename);
return;
}
fclose(temp_fp);
}
text_stream_fp = rr_iopen(text_stream_filename);
if (out_probs)
probs_stream_fp = rr_oopen(probs_stream_filename);
if (annotate)
annotation_fp = rr_oopen(annotation_filename);
if (out_oovs)
oov_fp = rr_oopen(oov_filename);
context = (id__t *) rr_malloc(sizeof(id__t)*(n-1));
sum_log_prob = 0.0;
total_words = 0;
excluded_unks = 0;
excluded_ccs = 0;
while (!rr_feof(text_stream_fp)) {
if (total_words > 0) {
if (total_words < n)
strcpy(prev_words[total_words-1],current_word);
else {
for (i=0;i<=n-3;i++)
strcpy(prev_words[i],prev_words[i+1]);
if (n>1)
strcpy(prev_words[n-2],current_word);
}
}
if (total_words < (n-1))
context_length = total_words;
else
context_length = n-1;
/* Fill context with right stuff */
if (total_words > (n-1)) {
for (i=0;i<=context_length-2;i++)
context[i] = context[i+1];
}
if (context_length != 0)
context[context_length-1] = short_current_id;
if (fscanf(text_stream_fp,"%s",current_word) != 1) {
if (!rr_feof(text_stream_fp)) {
printf("Error reading text file.\n");
return;
}
}
if (!rr_feof(text_stream_fp)) {
if (arpa_lm) {
sih_lookup(arpa_ng->vocab_ht,current_word,¤t_id);
if (arpa_ng->vocab_type == CLOSED_VOCAB && current_id == 0) {
found_unk_wrongly = 1;
printf("Error : %s is not in the vocabulary, and this is a closed \nvocabulary model.\n",current_word);
}
if (current_id > arpa_ng->vocab_size)
quit(-1,"Error : returned value from sih_lookup (%d) is too high.\n",context[i]);
else
short_current_id = current_id;
}else {
sih_lookup(ng->vocab_ht,current_word,¤t_id);
if (ng->vocab_type == CLOSED_VOCAB && current_id == 0) {
found_unk_wrongly = 1;
printf("Error : %s is not in the vocabulary, and this is a closed \nvocabulary model.\n",current_word);
}
if (current_id > ng->vocab_size)
quit(-1,"Error : returned value from sih_lookup (%d) is too high.\n",context[i]);
else
short_current_id = current_id;
}
if (!found_unk_wrongly) {
if (current_id == 0 && out_oovs)
fprintf(oov_fp,"%s\n",current_word);
if ((arpa_lm && (!(arpa_ng->context_cue[current_id])))
|| ((!arpa_lm) && (!(ng->context_cue[current_id])))) {
if (include_unks || current_id != 0) {
prob = calc_prob_of(short_current_id,
context,
context_length,
ng,
arpa_ng,
fb_list,
&bo_case,
&actual_context_length,
arpa_lm);
if (prob<= 0.0 || prob > 1.0) {
fprintf(stderr,"Warning : ");
if (short_current_id == 0)
fprintf(stderr,"P( <UNK> | ");
else
fprintf(stderr,"P( %s | ",current_word);
for (i=0;i<=actual_context_length-1;i++) {
if (context[i+context_length-actual_context_length] == 0)
fprintf(stderr,"<UNK> ");
else
fprintf(stderr,"%s ",prev_words[i]);
}
fprintf(stderr,") = %g logprob = %g \n ",prob,log(prob)/log(log_base));
fprintf(stderr,"bo_case == 0x%dx, actual_context_length == %d\n",
bo_case, actual_context_length);
}
if (annotate) {
if (short_current_id == 0)
fprintf(annotation_fp,"P( <UNK> | ");
else
fprintf(annotation_fp,"P( %s | ",current_word);
for (i=0;i<=actual_context_length-1;i++) {
if (context[i+context_length-actual_context_length] == 0)
fprintf(annotation_fp,"<UNK> ");
else {
if (arpa_lm)
fprintf(annotation_fp,"%s ",arpa_ng->vocab[context[i+context_length-actual_context_length]]);
else
fprintf(annotation_fp,"%s ",ng->vocab[context[i+context_length-actual_context_length]]);
}
}
fprintf(annotation_fp,") = %g logprob = %f bo_case = ",prob,log(prob)/log(log_base));
decode_bo_case(bo_case,actual_context_length,annotation_fp);
}
/* Calculate level to which we backed off */
for (i=actual_context_length-1;i>=0;i--) {
int four_raise_i = 1<<(2*i); /* PWP */
/*
* PWP: This was "if ((bo_case / (int) pow(3,i)) == 0)"
* but was getting a divide-by-zero error on an Alpha
* (it isn't clear to me why it should ever have done so)
* Anyway, it is much faster to do in base-4.
*/
if ((bo_case == 0) || ((bo_case / four_raise_i) == 0)) {
ngrams_hit[i+1]++;
i = -2;
}else
bo_case -= ((bo_case / four_raise_i) * four_raise_i);
}
if (i != -3)
ngrams_hit[0]++;
if (out_probs)
fprintf(probs_stream_fp,"%g\n",prob);
sum_log_prob += log10(prob);
}
if (current_id == 0 && !include_unks)
excluded_unks++;
}
else {
if (((!arpa_lm) && ng->context_cue[current_id]) ||
(arpa_lm && arpa_ng->context_cue[current_id]))
excluded_ccs++;
}
total_words++;
}
}
}
if (!found_unk_wrongly) { /* pow(x,y) = e**(y ln(x)) */
printf("Perplexity = %.2f, Entropy = %.2f bits\n",
exp(-sum_log_prob/(total_words-excluded_ccs-excluded_unks) *
log(10.0)),
(-sum_log_prob/(total_words-excluded_ccs-excluded_unks) *
log(10.0) / log(2.0)));
printf("Computation based on %d words.\n",
total_words-excluded_ccs-excluded_unks);
for(i=n;i>=1;i--) {
printf("Number of %d-grams hit = %d (%.2f%%)\n",i,ngrams_hit[i-1],
(float) 100*ngrams_hit[i-1]/(total_words-excluded_ccs-excluded_unks) );
}
printf("%d OOVs (%.2f%%) and %d context cues were removed from the calculation.\n",
excluded_unks,
(float) 100*excluded_unks/(total_words-excluded_ccs),excluded_ccs);
}
rr_iclose(text_stream_fp);
if (out_probs)
rr_oclose(probs_stream_fp);
if (annotate)
rr_oclose(annotation_fp);
if (out_oovs)
rr_oclose(oov_fp);
free (fb_list);
free (context);
free (ngrams_hit);
}
void main(int argc, char *argv[]) {
int verbosity;
int vocab_size;
int cutoff;
int num_recs;
int current_rec;
int num_above_threshold;
int num_to_output;
int i;
word_rec *records;
char temp_word[750];
flag gt_set;
flag top_set;
/* Process command line */
report_version(&argc,argv);
if (pc_flagarg( &argc, argv,"-help")) {
fprintf(stderr,"wfreq2vocab : Generate a vocabulary file from a word frequency file.\n");
fprintf(stderr,"Usage : wfreq2vocab [ -top 20000 | -gt 10]\n");
fprintf(stderr," [ -records %d ]\n",DEFAULT_MAX_RECORDS);
fprintf(stderr," [ -verbosity %d]\n",DEFAULT_VERBOSITY);
fprintf(stderr," < .wfreq > .vocab\n");
exit(1);
}
cutoff = pc_intarg( &argc, argv, "-gt",-1);
vocab_size = pc_intarg(&argc, argv, "-top",-1);
num_recs = pc_intarg(&argc, argv, "-records",DEFAULT_MAX_RECORDS);
verbosity = pc_intarg(&argc, argv, "-verbosity",DEFAULT_VERBOSITY);
pc_report_unk_args(&argc,argv,verbosity);
if (cutoff != -1) {
gt_set = 1;
}
else {
gt_set = 0;
cutoff = 0;
}
if (vocab_size != -1) {
top_set = 1;
}
else {
top_set = 0;
vocab_size = 0;
}
if (gt_set && top_set) {
quit(-1,"wfreq2vocab : Error : Can't use both the -top and the -gt options.\n");
}
if (!gt_set && !top_set) {
vocab_size = 20000;
}
if (gt_set) {
pc_message(verbosity,2,"wfreq2vocab : Will generate a vocabulary containing all words which\n occurred more that %d times. Reading wfreq stream from stdin...\n",cutoff);
}
else {
pc_message(verbosity,2,"wfreq2vocab : Will generate a vocabulary containing the most\n frequent %d words. Reading wfreq stream from stdin...\n",vocab_size);
}
records = (word_rec *) rr_malloc(sizeof(word_rec)*num_recs);
current_rec = 0;
num_above_threshold = 0;
while (!rr_feof(stdin)) {
if (scanf("%s %d",temp_word,&(records[current_rec].count)) != 2) {
if (!rr_feof(stdin)) {
quit(-1,"Error reading unigram counts from standard input.\n");
}
}
else {
records[current_rec].word = salloc(temp_word);
if (gt_set && records[current_rec].count > cutoff) {
num_above_threshold++;
}
current_rec++;
}
}
/* Sort records in descending order of count */
qsort((void*) records,(size_t) current_rec, sizeof(word_rec),sort_by_count);
if (gt_set) {
num_to_output = num_above_threshold;
}
else {
num_to_output = vocab_size;
}
if (current_rec<num_to_output) {
num_to_output = current_rec;
}
/* Now sort the relevant records alphabetically */
qsort((void*) records,(size_t) num_to_output, sizeof(word_rec),sort_alpha);
if (gt_set) {
pc_message(verbosity,2,"Size of vocabulary = %d\n",num_to_output);
}
if (num_to_output>65535) {
pc_message(verbosity,1,"Warning : Vocab size exceeds 65535. This will cause problems with \nother tools, since word id's are stored in 2 bytes.\n");
}
/* Print the vocab to stdout */
printf("## Vocab generated by v2 of the CMU-Cambridge Statistcal\n");
printf("## Language Modeling toolkit.\n");
printf("##\n");
printf("## Includes %d words ",num_to_output);
printf("##\n");
for (i=0;i<=num_to_output-1;i++) {
printf("%s\n",records[i].word);
}
pc_message(verbosity,0,"wfreq2vocab : Done.\n");
exit(0);
}
void main(int argc, char *argv[]) {
int i,j;
char *vocab_filename;
FILE *tempfile;
char tempfiles_directory[1000];
int vocab_size;
FILE *vocab_file;
int verbosity;
int buffer_size;
int position_in_buffer;
int number_of_tempfiles;
int max_files;
int fof_size;
unsigned short *buffer;
unsigned short *placeholder;
unsigned short *temp_ngram;
int temp_count;
char temp_word[500];
char temp_word2[500];
char *temp_file_root;
char *temp_file_ext;
char *host_name;
int proc_id;
struct utsname uname_info;
flag write_ascii;
/* Vocab hash table things */
struct hash_table vocabulary;
unsigned long hash_size;
unsigned long M;
tempfile = NULL; /* Just to prevent compilation warnings. */
report_version(&argc,argv);
verbosity = pc_intarg(&argc,argv,"-verbosity",DEFAULT_VERBOSITY);
/* Process command line */
if (pc_flagarg( &argc, argv,"-help") || argc==1) {
fprintf(stderr,"text2idngram - Convert a text stream to an id n-gram stream.\n");
fprintf(stderr,"Usage : text2idngram -vocab .vocab \n");
fprintf(stderr," [ -buffer 100 ]\n");
fprintf(stderr," [ -hash %d ]\n",DEFAULT_HASH_SIZE);
fprintf(stderr," [ -temp %s ]\n",DEFAULT_TEMP);
fprintf(stderr," [ -files %d ]\n",DEFAULT_MAX_FILES);
fprintf(stderr," [ -gzip | -compress ]\n");
fprintf(stderr," [ -verbosity %d ]\n",
DEFAULT_VERBOSITY);
fprintf(stderr," [ -n 3 ]\n");
fprintf(stderr," [ -write_ascii ]\n");
fprintf(stderr," [ -fof_size 10 ]\n");
exit(1);
}
pc_message(verbosity,2,"text2idngram\n");
n = pc_intarg( &argc, argv, "-n",DEFAULT_N);
placeholder = (unsigned short *) rr_malloc(sizeof(unsigned short)*n);
temp_ngram = (unsigned short *) rr_malloc(sizeof(unsigned short)*n);
hash_size = pc_intarg( &argc, argv, "-hash",DEFAULT_HASH_SIZE);
buffer_size = pc_intarg( &argc, argv, "-buffer",STD_MEM);
write_ascii = pc_flagarg(&argc,argv,"-write_ascii");
fof_size = pc_intarg(&argc,argv,"-fof_size",10);
max_files = pc_intarg( &argc, argv, "-files",DEFAULT_MAX_FILES);
vocab_filename = salloc(pc_stringarg( &argc, argv, "-vocab", "" ));
if (!strcmp("",vocab_filename)) {
quit(-1,"text2idngram : Error : Must specify a vocabulary file.\n");
}
strcpy(tempfiles_directory,pc_stringarg( &argc, argv, "-temp",
DEFAULT_TEMP));
if (pc_flagarg(&argc,argv,"-compress")) {
temp_file_ext = salloc(".Z");
}
else {
if (pc_flagarg(&argc,argv,"-gzip")) {
temp_file_ext = salloc(".gz");
}
else {
temp_file_ext = salloc("");
}
}
uname(&uname_info);
host_name = salloc(uname_info.nodename);
proc_id = getpid();
sprintf(temp_word,"%s%s.%d.",TEMP_FILE_ROOT,host_name,proc_id);
temp_file_root = salloc(temp_word);
pc_report_unk_args(&argc,argv,verbosity);
/* If the last charactor in the directory name isn't a / then add one. */
if (tempfiles_directory[strlen(tempfiles_directory)-1] != '/') {
strcat(tempfiles_directory,"/");
}
pc_message(verbosity,2,"Vocab : %s\n",vocab_filename);
pc_message(verbosity,2,"N-gram buffer size : %d\n",buffer_size);
pc_message(verbosity,2,"Hash table size : %d\n",hash_size);
pc_message(verbosity,2,"Temp directory : %s\n",tempfiles_directory);
pc_message(verbosity,2,"Max open files : %d\n",max_files);
pc_message(verbosity,2,"FOF size : %d\n",fof_size);
pc_message(verbosity,2,"n : %d\n",n);
buffer_size *= (1000000/(sizeof(unsigned short)*n));
/* Allocate memory for hash table */
fprintf(stderr,"Initialising hash table...\n");
M = nearest_prime(hash_size);
new_hashtable(&vocabulary,M);
/* Read in the vocabulary */
vocab_size = 0;
vocab_file = rr_iopen(vocab_filename);
pc_message(verbosity,2,"Reading vocabulary...\n");
while (fgets (temp_word, sizeof(temp_word),vocab_file)) {
if (strncmp(temp_word,"##",2)==0) continue;
sscanf (temp_word, "%s ",temp_word2);
/* Check for repeated words in the vocabulary */
if (index2(&vocabulary,temp_word2) != 0) {
fprintf(stderr,"======================================================\n");
fprintf(stderr,"WARNING: word %s is repeated in the vocabulary.\n",temp_word);
fprintf(stderr,"=======================================================\n");
}
if (strncmp(temp_word,"#",1)==0) {
fprintf(stderr,"\n\n===========================================================\n");
fprintf(stderr,":\nWARNING: line assumed NOT a comment:\n");
fprintf(stderr, ">>> %s <<<\n",temp_word);
fprintf(stderr, " '%s' will be included in the vocabulary.\n",temp_word2);
fprintf(stderr, " (comments must start with '##')\n");
fprintf(stderr,"===========================================================\n\n");
}
vocab_size++;
add_to_hashtable(&vocabulary,hash(temp_word2,M),temp_word2,vocab_size);
}
if (vocab_size > MAX_VOCAB_SIZE) {
quit(-1,"text2idngram : Error : Vocabulary size exceeds maximum.\n");
}
pc_message(verbosity,2,"Allocating memory for the n-gram buffer...\n");
buffer=(unsigned short*) rr_malloc(n*(buffer_size+1)*sizeof(unsigned short));
number_of_tempfiles = 0;
/* Read text into buffer */
/* Read in the first ngram */
position_in_buffer = 0;
for (i=0;i<=n-1;i++) {
get_word(stdin,temp_word);
add_to_buffer(index2(&vocabulary,temp_word),0,i,buffer);
}
while (!rr_feof(stdin)) {
/* Fill up the buffer */
pc_message(verbosity,2,"Reading text into the n-gram buffer...\n");
pc_message(verbosity,2,"20,000 n-grams processed for each \".\", 1,000,000 for each line.\n");
while ((position_in_buffer<buffer_size) && (!rr_feof(stdin))) {
position_in_buffer++;
if (position_in_buffer % 20000 == 0) {
if (position_in_buffer % 1000000 == 0) {
pc_message(verbosity,2,".\n");
}
else {
pc_message(verbosity,2,".");
}
}
for (i=1;i<=n-1;i++) {
add_to_buffer(buffer_contents(position_in_buffer-1,i,buffer),
position_in_buffer,i-1,buffer);
}
if (get_word(stdin,temp_word) == 1) {
add_to_buffer(index2(&vocabulary,temp_word),position_in_buffer,
n-1,buffer);
}
}
for (i=0;i<=n-1;i++) {
placeholder[i] = buffer_contents(position_in_buffer,i,buffer);
}
/* Sort buffer */
pc_message(verbosity,2,"\nSorting n-grams...\n");
qsort((void*) buffer,(size_t) position_in_buffer,
n*sizeof(unsigned short),compare_ngrams);
/* Output the buffer to temporary BINARY file */
number_of_tempfiles++;
sprintf(temp_word,"%s%s%hu%s",tempfiles_directory,temp_file_root,
number_of_tempfiles,temp_file_ext);
pc_message(verbosity,2,"Writing sorted n-grams to temporary file %s\n",
temp_word);
tempfile = rr_oopen(temp_word);
for (i=0;i<=n-1;i++) {
temp_ngram[i] = buffer_contents(0,i,buffer);
if (temp_ngram[i] > MAX_VOCAB_SIZE) {
quit(-1,"Invalid trigram in buffer.\nAborting");
}
}
temp_count = 1;
for (i=1;i<=position_in_buffer;i++) {
if (!compare_ngrams(temp_ngram,&buffer[i*n])) {
temp_count++;
}
else {
for (j=0;j<=n-1;j++) {
rr_fwrite(&temp_ngram[j],sizeof(unsigned short),1,
tempfile,"temporary n-gram ids");
temp_ngram[j] = buffer_contents(i,j,buffer);
}
rr_fwrite(&temp_count,sizeof(int),1,tempfile,
"temporary n-gram counts");
temp_count = 1;
}
}
rr_oclose(tempfile);
for (i=0;i<=n-1;i++) {
add_to_buffer(placeholder[i],0,i,buffer);
}
position_in_buffer = 0;
}
/* Merge the temporary files, and output the result to standard output */
pc_message(verbosity,2,"Merging temporary files...\n");
merge_tempfiles(1,
number_of_tempfiles,
temp_file_root,
temp_file_ext,
max_files,
tempfiles_directory,
stdout,
write_ascii,
fof_size);
pc_message(verbosity,0,"text2idngram : Done.\n");
exit(0);
}
int oe_03_main (int argc, char **argv) {
flag first_ngram;
int n;
fof_sz_t fof_size;
flag is_ascii;
int verbosity;
fof_t **fof_array;
ngram_sz_t *num_kgrams;
ngram current_ngram;
ngram previous_ngram;
count_t *ng_count;
int pos_of_novelty;
int nlines;
int i;
report_version(&argc,argv);
if (argc == 1 || pc_flagarg(&argc, argv,"-help")) {
oe_04_help_message();
exit(1);
}
is_ascii = pc_flagarg(&argc, argv,"-ascii_input");
n = pc_intarg(&argc, argv,"-n",3);
fof_size = pc_intarg(&argc, argv,"-fof_size",50);
verbosity = pc_intarg(&argc, argv,"-verbosity",DEFAULT_VERBOSITY);
pc_report_unk_args(&argc,argv,verbosity);
pc_message(verbosity,2,"n = %d\n",n);
pc_message(verbosity,2,"fof_size = %d\n",fof_size);
current_ngram.n = n;
previous_ngram.n = n;
pos_of_novelty = n;
fof_array = (fof_t **) rr_malloc(sizeof(fof_t *) * (n-1));
for (i=0;i<=n-2;i++)
fof_array[i] = (fof_t *) rr_calloc(fof_size+1,sizeof(fof_t));
num_kgrams = (ngram_sz_t *) rr_calloc(n-1,sizeof(ngram_sz_t));
ng_count = (count_t *) rr_calloc(n-1,sizeof(count_t));
current_ngram.id_array = (id__t *) rr_calloc(n,sizeof(id__t));
previous_ngram.id_array = (id__t *) rr_calloc(n,sizeof(id__t));
pc_message(verbosity,2,"Processing id n-gram file.\n");
pc_message(verbosity,2,"20,000 n-grams processed for each \".\", 1,000,000 for each line.\n");
nlines = 0;
first_ngram = 1;
while (!rr_feof(stdin)) {
if (!first_ngram)
ngram_copy(&previous_ngram,¤t_ngram,n);
if (get_ngram(stdin,¤t_ngram,is_ascii)) {
nlines++;
show_idngram_nlines(nlines, verbosity);
/* Test for where this ngram differs from last - do we have an
out-of-order ngram? */
if (!first_ngram)
pos_of_novelty = ngram_find_pos_of_novelty(¤t_ngram,&previous_ngram,n,nlines);
else
pos_of_novelty = 0;
/* Add new N-gram */
num_kgrams[n-2]++;
if (current_ngram.count <= fof_size)
fof_array[n-2][current_ngram.count]++;
if (!first_ngram) {
for (i=n-2;i>=MAX(1,pos_of_novelty);i--) {
num_kgrams[i-1]++;
if (ng_count[i-1] <= fof_size)
fof_array[i-1][ng_count[i-1]]++;
ng_count[i-1] = current_ngram.count;
}
} else {
for (i=n-2;i>=MAX(1,pos_of_novelty);i--)
ng_count[i-1] = current_ngram.count;
}
for (i=0;i<=pos_of_novelty-2;i++)
ng_count[i] += current_ngram.count;
if (first_ngram)
first_ngram = 0;
}
}
/* Process last ngram */
for (i=n-2;i>=MAX(1,pos_of_novelty);i--) {
num_kgrams[i-1]++;
if (ng_count[i-1] <= fof_size) {
fof_array[i-1][ng_count[i-1]]++;
}
ng_count[i-1] = current_ngram.count;
}
#import "OpenEarsStaticAnalysisToggle.h"
#ifdef STATICANALYZEDEPENDENCIES
#define __clang_analyzer__ 1
#endif
#if !defined(__clang_analyzer__) || defined(STATICANALYZEDEPENDENCIES)
#undef __clang_analyzer__
for (i=0;i<=pos_of_novelty-2;i++)
ng_count[i] += current_ngram.count;
display_fof_array(num_kgrams,fof_array,fof_size,stderr, n);
#endif
pc_message(verbosity,0,"idngram2stats : Done.\n");
exit(0);
}
int main (int argc, char **argv) {
int n;
int verbosity;
int max_files;
int max_words;
int max_chars;
int current_word;
int current_char;
int start_char; /* start boundary (possibly > than 0) */
int no_of_spaces;
int pos_in_string;
int i;
char *current_string;
char current_temp_filename[500];
int current_file_number;
FILE *temp_file;
flag text_buffer_full;
char *text_buffer;
char **pointers;
char current_ngram[500];
int current_count;
int counter;
char temp_directory[1000];
char *temp_file_ext;
flag words_set;
flag chars_set;
/* Process command line */
verbosity = pc_intarg(&argc, argv,"-verbosity",DEFAULT_VERBOSITY);
pc_message(verbosity,2,"text2wngram\n");
report_version(&argc,argv);
if (pc_flagarg( &argc, argv,"-help")) {
help_message();
exit(1);
}
n = pc_intarg(&argc, argv,"-n",DEFAULT_N);
/* max_words = pc_intarg(&argc, argv,"-words",STD_MEM*1000000/11);
max_chars = pc_intarg(&argc, argv,"-chars",STD_MEM*7000000/11); */
max_words = pc_intarg(&argc, argv,"-words",-1);
max_chars = pc_intarg(&argc, argv,"-chars",-1);
if (max_words == -1) {
words_set = 0;
max_words = STD_MEM*1000000/11;
}else
words_set = 1;
if (max_chars == -1) {
chars_set = 0;
max_chars = STD_MEM*7000000/11;
}else
chars_set = 1;
max_files = pc_intarg(&argc, argv,"-files",DEFAULT_MAX_FILES);
if (pc_flagarg(&argc,argv,"-compress"))
temp_file_ext = salloc(".Z");
else {
if (pc_flagarg(&argc,argv,"-gzip"))
temp_file_ext = salloc(".gz");
else
temp_file_ext = salloc("");
}
strcpy(temp_directory, "cmuclmtk-XXXXXX");
if (mkdtemp(temp_directory) == NULL) {
quit(-1, "Failed to create temporary folder: %s\n", strerror(errno));
}
pc_report_unk_args(&argc,argv,verbosity);
if (words_set && !chars_set)
max_chars = max_words * 7;
if (!words_set && chars_set)
max_words = max_chars / 7;
/* If the last charactor in the directory name isn't a / then add one. */
pc_message(verbosity,2,"n = %d\n",n);
pc_message(verbosity,2,"Number of words in buffer = %d\n",max_words);
pc_message(verbosity,2,"Number of chars in buffer = %d\n",max_chars);
pc_message(verbosity,2,"Max number of files open at once = %d\n",max_files);
pc_message(verbosity,2,"Temporary directory = %s\n",temp_directory);
/* Allocate memory for the buffers */
text_buffer = (char *) rr_malloc(sizeof(char)*max_chars);
pc_message(verbosity,2,"Allocated %d bytes to text buffer.\n",
sizeof(char)*max_chars);
pointers = (char **) rr_malloc(sizeof(char *)*max_words);
pc_message(verbosity,2,"Allocated %d bytes to pointer array.\n",
sizeof(char *)*max_words);
current_file_number = 0;
current_word = 1;
start_char = 0;
current_char = 0;
counter = 0;
pointers[0] = text_buffer;
while (!feof(stdin)) {
current_file_number++;
/* Read text into buffer */
pc_message(verbosity,2,"Reading text into buffer...\n");
pc_message(verbosity,2,"Reading text into the n-gram buffer...\n");
pc_message(verbosity,2,"20,000 words processed for each \".\", 1,000,000 for each line.\n");
pointers[0] = text_buffer;
while ((!rr_feof(stdin)) &&
(current_word < max_words) &&
(current_char < max_chars)) {
text_buffer[current_char] = getchar();
if (text_buffer[current_char] == '\n' ||
text_buffer[current_char] == '\t' ) {
text_buffer[current_char] = ' ';
}
if (text_buffer[current_char] == ' ') {
if (current_char > start_char) {
if (text_buffer[current_char-1] == ' ') {
current_word--;
current_char--;
}
pointers[current_word] = &(text_buffer[current_char+1]);
current_word++;
counter++;
if (counter % 20000 == 0) {
if (counter % 1000000 == 0)
pc_message(verbosity,2,"\n");
else
pc_message(verbosity,2,".");
}
}
}
if (text_buffer[current_char] != ' ' || current_char > start_char)
current_char++;
}
text_buffer[current_char]='\0';
if (current_word == max_words || rr_feof(stdin)) {
for (i=current_char+1;i<=max_chars-1;i++)
text_buffer[i] = ' ';
text_buffer_full = 0;
}else
text_buffer_full = 1;
/* Sort buffer */
pc_message(verbosity,2,"\nSorting pointer array...\n");
qsort((void *) pointers,(size_t) current_word-n,sizeof(char *),cmp_strings);
/* Write out temporary file */
sprintf(current_temp_filename,"%s/%hu%s",temp_directory, current_file_number, temp_file_ext);
pc_message(verbosity,2,"Writing out temporary file %s...\n",current_temp_filename);
temp_file = rr_oopen(current_temp_filename);
text_buffer[current_char] = ' ';
current_count = 0;
strcpy(current_ngram,"");
for (i = 0; i <= current_word-n; i++) {
current_string = pointers[i];
/* Find the nth space */
no_of_spaces = 0;
pos_in_string = 0;
while (no_of_spaces < n) {
if (current_string[pos_in_string] == ' ')
no_of_spaces++;
pos_in_string++;
}
if (!strncmp(current_string,current_ngram,pos_in_string))
current_count++;
else {
if (strcmp(current_ngram,""))
if (fprintf(temp_file,"%s %d\n",current_ngram,current_count) < 0)
quit(-1,"Error writing to temporary file %s\n",current_temp_filename);
current_count = 1;
strncpy(current_ngram,current_string,pos_in_string);
current_ngram[pos_in_string] = '\0';
}
}
rr_oclose(temp_file);
/* Move the last n-1 words to the beginning of the buffer, and set
correct current_word and current_char things */
strcpy(text_buffer,pointers[current_word-n]);
pointers[0]=text_buffer;
/* Find the (n-1)th space */
no_of_spaces=0;
pos_in_string=0;
if (!text_buffer_full){
while (no_of_spaces<(n-1)) {
if (pointers[0][pos_in_string]==' ') {
no_of_spaces++;
pointers[no_of_spaces] = &pointers[0][pos_in_string+1];
}
pos_in_string++;
}
}else {
while (no_of_spaces<n) {
if (pointers[0][pos_in_string]==' ') {
no_of_spaces++;
pointers[no_of_spaces] = &pointers[0][pos_in_string+1];
}
pos_in_string++;
}
pos_in_string--;
}
current_char = pos_in_string;
current_word = n;
/* mark boundary beyond which counting pass cannot backup */
start_char = current_char;
}
/* Merge temporary files */
pc_message(verbosity,2,"Merging temporary files...\n");
merge_tempfiles(1,
current_file_number,
temp_directory,
temp_file_ext,
max_files,
stdout,
n,
verbosity);
rmdir(temp_directory);
pc_message(verbosity,0,"text2wngram : Done.\n");
return 0;
}
int main(int argc, char *argv[]) {
int verbosity;
int vocab_size;
FILE *vocab_file;
int buffer_size;
flag write_ascii;
int max_files;
int number_of_tempfiles;
char *vocab_filename;
char *idngram_filename;
char temp_word[MAX_WORD_LENGTH];
char temp_word2[MAX_WORD_LENGTH];
char temp_word3[MAX_WORD_LENGTH];
flag contains_unks;
int position_in_buffer;
FILE *outfile;
FILE *tempfile;
FILE *non_unk_fp;
ngram_rec *buffer;
flag same_ngram;
int i;
int j;
int fof_size;
int size_of_rec;
char temp_directory[1000];
char *temp_file_ext;
/* Vocab hash table things */
struct idngram_hash_table vocabulary;
unsigned long hash_size;
unsigned long M;
wordid_t *current_ngram;
int current_count;
wordid_t *sort_ngram;
int sort_count;
/* Process command line */
report_version(&argc,argv);
if (argc == 1 || pc_flagarg(&argc, argv,"-help")) {
/* Display help message */
help_message();
exit(1);
}
n = pc_intarg( &argc, argv, "-n",DEFAULT_N);
hash_size = pc_intarg( &argc, argv, "-hash",DEFAULT_HASH_SIZE);
buffer_size = pc_intarg( &argc, argv, "-buffer",STD_MEM);
write_ascii = pc_flagarg(&argc,argv,"-write_ascii");
verbosity = pc_intarg(&argc,argv,"-verbosity",DEFAULT_VERBOSITY);
max_files = pc_intarg( &argc, argv, "-files",DEFAULT_MAX_FILES);
fof_size = pc_intarg(&argc,argv,"-fof_size",10);
vocab_filename = salloc(pc_stringarg( &argc, argv, "-vocab", "" ));
idngram_filename = salloc(pc_stringarg( &argc, argv, "-idngram", "" ));
if (!strcmp("",vocab_filename))
quit(-1,"Error : Must specify a vocabulary file.\n");
if (!strcmp("",idngram_filename))
quit(-1,"text2idngram : Error : Must specify idngram file.\n");
if (pc_flagarg(&argc,argv,"-compress"))
temp_file_ext = salloc(".Z");
else {
if (pc_flagarg(&argc,argv,"-gzip"))
temp_file_ext = salloc(".gz");
else
temp_file_ext = salloc("");
}
strcpy(temp_directory, "cmuclmtk-XXXXXX");
if (mkdtemp(temp_directory) == NULL) {
quit(-1, "Failed to create temporary folder: %s\n", strerror(errno));
}
pc_report_unk_args(&argc,argv,verbosity);
outfile = rr_fopen(idngram_filename,"wb");
pc_message(verbosity,2,"Vocab : %s\n",vocab_filename);
pc_message(verbosity,2,"Output idngram : %s\n",idngram_filename);
pc_message(verbosity,2,"Buffer size : %d\n",buffer_size);
pc_message(verbosity,2,"Hash table size : %d\n",hash_size);
pc_message(verbosity,2,"Max open files : %d\n",max_files);
pc_message(verbosity,2,"n : %d\n",n);
pc_message(verbosity,2,"FOF size : %d\n",fof_size);
size_of_rec = (sizeof(wordid_t) * n) + 16 - (( n* sizeof(wordid_t)) % 16);
buffer_size *= (1000000/((sizeof(ngram_rec) + size_of_rec)));
fprintf(stderr,"buffer size = %d\n",buffer_size);
/* Allocate memory for hash table */
fprintf(stderr,"Initialising hash table...\n");
M = nearest_prime(hash_size);
new_idngram_hashtable(&vocabulary,M);
/* Read in the vocabulary */
vocab_size = 0;
vocab_file = rr_iopen(vocab_filename);
pc_message(verbosity,2,"Reading vocabulary...\n");
while (fgets (temp_word, sizeof(temp_word),vocab_file)) {
if (strncmp(temp_word,"##",2)==0) continue;
sscanf (temp_word, "%s ",temp_word2);
/* Check for vocabulary order */
if (vocab_size > 0 && strcmp(temp_word2,temp_word3)<0)
quit(-1,"wngram2idngram : Error : Vocabulary is not alphabetically ordered.\n");
/* Check for repeated words in the vocabulary */
if (index2(&vocabulary,temp_word2) != 0)
warn_on_repeated_words(temp_word);
warn_on_wrong_vocab_comments(temp_word);
vocab_size++;
add_to_idngram_hashtable(&vocabulary,idngram_hash(temp_word2,M),temp_word2,vocab_size);
strcpy(temp_word3,temp_word2);
}
if (vocab_size > MAX_VOCAB_SIZE)
quit(-1,"Error : Vocabulary size exceeds maximum.\n");
pc_message(verbosity,2,"Allocating memory for the buffer...\n");
buffer=(ngram_rec *) rr_malloc((buffer_size+1)*sizeof(ngram_rec));
for (i=0;i<=buffer_size;i++)
buffer[i].word = (wordid_t *) rr_malloc(n*sizeof(wordid_t));
/* Open the "non-OOV" tempfile */
sprintf(temp_word, "%s/1%s", temp_directory, temp_file_ext);
non_unk_fp = rr_fopen(temp_word,"w");
pc_message(verbosity,2,"Writing non-OOV counts to temporary file %s\n",
temp_word);
number_of_tempfiles = 1;
current_ngram = (wordid_t *) rr_malloc(n*sizeof(wordid_t));
sort_ngram = (wordid_t *) rr_malloc(n*sizeof(wordid_t));
/* Read text into buffer */
position_in_buffer = 0;
while (!rr_feof(stdin)) {
for (i=0;i<=n-1;i++) {
get_word(stdin,temp_word);
current_ngram[i]=index2(&vocabulary,temp_word);
}
if (scanf("%d",¤t_count) != 1)
if (!rr_feof(stdin))
quit(-1,"Error reading n-gram count from stdin.\n");
if (!rr_feof(stdin)) {
contains_unks = 0;
for (i=0;i<=n-1;i++) {
if (!current_ngram[i])
contains_unks = 1;
}
if (contains_unks) {
/* Write to buffer */
position_in_buffer++;
if (position_in_buffer >= buffer_size) {
/* Sort buffer */
pc_message(verbosity,2,
"Sorting n-grams which include an OOV word...\n");
qsort((void*) buffer,(size_t) position_in_buffer,
sizeof(ngram_rec),compare_ngrams2);
pc_message(verbosity,2,"Done.\n");
/* Write buffer to temporary file */
number_of_tempfiles++;
sprintf(temp_word,"%s/%hu%s", temp_directory,
number_of_tempfiles,temp_file_ext);
pc_message(verbosity,2,
"Writing sorted OOV-counts buffer to temporary file %s\n",
temp_word);
tempfile = rr_fopen(temp_word,"w");
for (i=0;i<=n-1;i++)
sort_ngram[i] = buffer[0].word[i];
sort_count = buffer[0].count;
for (i=0;i<=position_in_buffer-2;i++) {
same_ngram = 1;
for (j=n-1;j>=0;j--) {
if (buffer[i].word[j] != sort_ngram[j]) {
same_ngram = 0;
j = -1;
}
}
if (same_ngram)
sort_count += buffer[i].count;
else {
for (j=0;j<=n-1;j++) {
rr_fwrite((char*)&sort_ngram[j],sizeof(wordid_t),1,
tempfile,"temporary n-gram ids");
sort_ngram[j] = buffer[i].word[j];
}
rr_fwrite((char*)&sort_count,sizeof(int),1,tempfile,
"temporary n-gram counts");
sort_count = buffer[i].count;
}
}
for (j=0;j<=n-1;j++)
rr_fwrite((char*)&sort_ngram[j],sizeof(wordid_t),1,
tempfile,"temporary n-gram ids");
rr_fwrite((char*)&sort_count,sizeof(int),1,tempfile,
"temporary n-gram counts");
rr_oclose(tempfile);
position_in_buffer = 1;
}
for (i=0;i<=n-1;i++)
buffer[position_in_buffer-1].word[i] = current_ngram[i];
buffer[position_in_buffer-1].count = current_count;
}else {
/* Write to temporary file */
for (i=0;i<=n-1;i++)
rr_fwrite((char*)¤t_ngram[i],sizeof(wordid_t),1,
non_unk_fp,"temporary n-gram ids");
rr_fwrite((char*)¤t_count,sizeof(int),1,non_unk_fp,
"temporary n-gram counts");
}
}
}
if (position_in_buffer > 0) {
/* Only do this bit if we have actually seen some OOVs */
/* Sort final buffer */
pc_message(verbosity,2,"Sorting final buffer...\n");
qsort((void*) buffer,(size_t) position_in_buffer,
sizeof(ngram_rec),compare_ngrams2);
/* Write final buffer */
number_of_tempfiles++;
sprintf(temp_word,"%s/%hu%s", temp_directory,
number_of_tempfiles,temp_file_ext);
pc_message(verbosity,2,"Writing sorted buffer to temporary file %s\n", temp_word);
tempfile = rr_fopen(temp_word,"w");
for (i=0;i<=n-1;i++)
sort_ngram[i] = buffer[0].word[i];
sort_count = buffer[0].count;
for (i=1;i<=position_in_buffer-1;i++) {
same_ngram = 1;
for (j=n-1;j>=0;j--) {
if (buffer[i].word[j] != sort_ngram[j]) {
same_ngram = 0;
j = -1;
}
}
if (same_ngram)
sort_count += buffer[i].count;
else {
for (j=0;j<=n-1;j++) {
rr_fwrite((char*)&sort_ngram[j],sizeof(wordid_t),1,
tempfile,"temporary n-gram ids");
sort_ngram[j] = buffer[i].word[j];
}
rr_fwrite((char*)&sort_count,sizeof(int),1,tempfile,
"temporary n-gram counts");
sort_count = buffer[i].count;
}
}
for (j=0;j<=n-1;j++)
rr_fwrite((char*)&sort_ngram[j],sizeof(wordid_t),1,
tempfile,"temporary n-gram ids");
rr_fwrite((char*)&sort_count,sizeof(int),1,tempfile,
"temporary n-gram counts");
fclose(tempfile);
}
/* Merge the temporary files, and output the result */
fclose(non_unk_fp);
pc_message(verbosity,2,"Merging temporary files...\n");
merge_idngramfiles(1,
number_of_tempfiles,
temp_directory,
temp_file_ext,
max_files,
outfile,
write_ascii,
fof_size,
n);
fclose(outfile);
rmdir(temp_directory);
pc_message(verbosity,0,"wngram2idngram : Done.\n");
return 0;
}
void main(int argc, char *argv[]) {
int verbosity;
int n;
int m;
int i;
int input_type;
int storage_type;
unsigned short *current_ngram_int;
unsigned short *previous_ngram_int;
char **current_ngram_text;
char **previous_ngram_text;
int current_count;
int running_total;
flag same;
flag first_one;
flag got_to_eof;
running_total = 0;
report_version(&argc,argv);
if (pc_flagarg( &argc, argv,"-help") || argc==1) {
fprintf(stderr,"ngram2mgram - Convert an n-gram file to an m-gram file, where m<n\n");
fprintf(stderr,"Usage : ngram2mgram -n N -m M\n");
fprintf(stderr," [ -binary | -ascii | -words ]\n");
fprintf(stderr," < .ngram > .mgram\n");
exit(1);
}
n = pc_intarg( &argc, argv,"-n",0);
m = pc_intarg( &argc, argv,"-m",0);
verbosity = pc_intarg(&argc,argv,"-verbosity",DEFAULT_VERBOSITY);
input_type = 0;
if (pc_flagarg( &argc, argv,"-binary")) {
input_type = BINARY;
}
if (pc_flagarg( &argc, argv,"-ascii")) {
if (input_type != 0) {
quit(-1,"Error : more than one file format specified.\n");
}
input_type = ASCII;
}
if (pc_flagarg( &argc, argv,"-words")) {
if (input_type != 0) {
quit(-1,"Error : more than one file format specified.\n");
}
input_type = WORDS;
}
if (input_type == 0) {
pc_message(verbosity,2,"Warning : no input type specified. Defaulting to binary.\n");
input_type = BINARY;
}
if (n == 0) {
quit(-1,"Must specify a value for n. Use the -n switch.\n");
}
if (m == 0) {
quit(-1,"Must specify a value for m. Use the -m switch.\n");
}
if (n<=m) {
quit(-1,"n must be greater than m.\n");
}
pc_report_unk_args(&argc,argv,verbosity);
if (input_type == BINARY || input_type == ASCII) {
storage_type = NUMERIC;
}
else {
storage_type = ALPHA;
}
if (storage_type == NUMERIC) {
current_ngram_int = (unsigned short *)
rr_malloc(n*sizeof(unsigned short));
previous_ngram_int = (unsigned short *)
rr_malloc(n*sizeof(unsigned short));
/* And to prevent compiler warnings ... */
current_ngram_text = NULL;
previous_ngram_text = NULL;
}
else {
current_ngram_text = (char **) rr_malloc(n*sizeof(char *));
previous_ngram_text = (char **) rr_malloc(n*sizeof(char *));
for (i=0;i<=n-1;i++) {
current_ngram_text[i] = (char *) rr_malloc(MAX_WORD_LENGTH*sizeof(char));
previous_ngram_text[i] = (char *) rr_malloc(MAX_WORD_LENGTH*sizeof(char));
}
/* And to prevent compiler warnings ... */
current_ngram_int = NULL;
previous_ngram_int = NULL;
}
got_to_eof = 0;
first_one = 1;
while (!rr_feof(stdin)) {
/* Store previous n-gram */
if (!first_one) {
if (storage_type == NUMERIC) {
for (i=0;i<=n-1;i++) {
previous_ngram_int[i] = current_ngram_int[i];
}
}
else {
for (i=0;i<=n-1;i++) {
strcpy(previous_ngram_text[i],current_ngram_text[i]);
}
}
}
/* Read new n-gram */
switch(input_type) {
case BINARY:
for (i=0;i<=n-1;i++) {
rr_fread(¤t_ngram_int[i],sizeof(id__t),1,stdin,
"from id_ngrams at stdin",0);
}
rr_fread(¤t_count,sizeof(count_t),1,stdin,
"from id_ngrams file at stdin",0);
break;
case ASCII:
for (i=0;i<=n-1;i++) {
if (fscanf(stdin,"%hu",¤t_ngram_int[i]) != 1) {
if (!rr_feof(stdin)) {
quit(-1,"Error reading id_ngram.\n");
}
else {
got_to_eof = 1;
}
}
}
if (fscanf(stdin,"%d",¤t_count) != 1) {
if (!rr_feof(stdin)) {
quit(-1,"Error reading id_ngram.\n");
}
else {
got_to_eof = 1;
}
}
break;
case WORDS:
for (i=0;i<=n-1;i++) {
if (fscanf(stdin,"%s",current_ngram_text[i]) != 1) {
if (!rr_feof(stdin)) {
quit(-1,"Error reading id_ngram.\n");
}
else {
got_to_eof = 1;
}
}
}
if (fscanf(stdin,"%d",¤t_count) != 1) {
if (!rr_feof(stdin)) {
quit(-1,"Error reading id_ngram.\n");
}
else {
got_to_eof = 1;
}
}
break;
}
if (!got_to_eof) {
/* Check for correct sorting */
if (!first_one) {
switch(storage_type) {
case NUMERIC:
for (i=0;i<=n-1;i++) {
if (current_ngram_int[i]<previous_ngram_int[i]) {
quit(-1,"Error : ngrams not correctly sorted.\n");
}
else {
if (current_ngram_int[i]>previous_ngram_int[i]) {
i=n;
}
}
}
break;
case ALPHA:
for (i=0;i<=n-1;i++) {
if (strcmp(current_ngram_text[i],previous_ngram_text[i])<0) {
quit(-1,"Error : ngrams not correctly sorted.\n");
}
else {
if (strcmp(current_ngram_text[i],previous_ngram_text[i])>0) {
i=n;
}
}
}
break;
}
}
/* Compare this m-gram with previous m-gram */
if (!first_one) {
switch(storage_type) {
case NUMERIC:
same = 1;
for (i=0;i<=m-1;i++) {
if (current_ngram_int[i] != previous_ngram_int[i]) {
same = 0;
}
}
if (same) {
running_total += current_count;
}
else {
if (input_type == ASCII) {
for (i=0;i<=m-1;i++) {
printf("%d ",previous_ngram_int[i]);
}
printf("%d\n",running_total);
}
else {
for (i=0;i<=m-1;i++) {
rr_fwrite(&previous_ngram_int[i],sizeof(id__t),1,stdout,
"to id_ngrams at stdout");
}
rr_fwrite(&running_total,sizeof(count_t),1,stdout,
"to id n-grams at stdout");
}
running_total = current_count;
}
break;
case ALPHA:
same = 1;
for (i=0;i<=m-1;i++) {
if (strcmp(current_ngram_text[i],previous_ngram_text[i])) {
same = 0;
}
}
if (same) {
running_total += current_count;
}
else {
for (i=0;i<=m-1;i++) {
printf("%s ",previous_ngram_text[i]);
}
printf("%d\n",running_total);
running_total = current_count;
}
break;
}
}
else {
running_total = current_count;
}
first_one = 0;
}
}
/* Write out final m-gram */
switch(input_type) {
case BINARY:
break;
case ASCII:
for (i=0;i<=m-1;i++) {
printf("%d ",previous_ngram_int[i]);
}
printf("%d\n",running_total);
break;
case WORDS:
for (i=0;i<=m-1;i++) {
printf("%s ",previous_ngram_text[i]);
}
printf("%d\n",running_total);
break;
}
pc_message(verbosity,0,"ngram2mgram : Done.\n");
exit(0);
}
void merge_idngramfiles (int start_file,
int end_file,
char *temp_file_root,
char *temp_file_ext,
int max_files,
FILE *outfile,
flag write_ascii,
int fof_size,
int n_order) {
FILE *new_temp_file;
char temp_string[1000];
char *new_temp_filename;
FILE **temp_file;
char **temp_filename;
wordid_t **current_ngram;
wordid_t *smallest_ngram;
wordid_t *previous_ngram;
int *current_ngram_count;
flag *finished;
flag all_finished;
int temp_count;
int i,j;
flag first_ngram;
fof_t **fof_array;
ngram_sz_t *num_kgrams;
int *ng_count;
int pos_of_novelty;
n = n_order;
pos_of_novelty = n; /* Simply for warning-free compilation */
num_kgrams = (ngram_sz_t *) rr_calloc(n-1,sizeof(ngram_sz_t));
ng_count = (int *) rr_calloc(n-1,sizeof(int));
first_ngram = 1;
previous_ngram = (wordid_t *) rr_calloc(n,sizeof(wordid_t));
temp_file = (FILE **) rr_malloc(sizeof(FILE *) * (end_file-start_file+1));
temp_filename = (char **) rr_malloc(sizeof(char *) *
(end_file-start_file+1));
/* should change to 2d array*/
current_ngram = (wordid_t **) rr_malloc(sizeof(wordid_t *) *
(end_file-start_file+1));
for (i=0;i<=end_file-start_file;i++)
current_ngram[i] = (wordid_t *) rr_malloc(sizeof(wordid_t)*n);
current_ngram_count = (int *) rr_malloc(sizeof(int)*(end_file-start_file+1));
finished = (flag *) rr_malloc(sizeof(flag)*(end_file-start_file+1));
smallest_ngram = (wordid_t *) rr_malloc(sizeof(wordid_t)*n);
/* should change to 2d array*/
fof_array = (fof_t **) rr_malloc(sizeof(fof_t *)*(n-1));
for (i=0;i<=n-2;i++)
fof_array[i] = (fof_t *) rr_calloc(fof_size+1,sizeof(fof_t));
if (end_file-start_file+1 > max_files) {
sprintf(temp_string,"%s/%hu%s",temp_file_root,
end_file+1,temp_file_ext);
new_temp_filename = salloc(temp_string);
new_temp_file = rr_oopen(new_temp_filename);
merge_tempfiles(start_file,start_file+max_files-1,
temp_file_root,temp_file_ext,max_files,
new_temp_file,write_ascii,0);
merge_tempfiles(start_file+max_files,end_file+1,
temp_file_root,temp_file_ext,max_files,
outfile,write_ascii,0);
}else {
/* Open all the temp files for reading */
for (i=0;i<=end_file-start_file;i++) {
sprintf(temp_string,"%s/%hu%s",temp_file_root,
i+start_file,temp_file_ext);
temp_filename[i] = salloc(temp_string);
temp_file[i] = rr_iopen(temp_filename[i]);
}
/* Now go through the files simultaneously, and write out the appropriate
ngram counts to the output file. */
for (i=end_file-start_file;i>=0;i--) {
finished[i] = 0;
if (!rr_feof(temp_file[i])) {
for (j=0;j<=n-1;j++) {
rr_fread((char*) ¤t_ngram[i][j], sizeof(wordid_t),1,
temp_file[i],"temporary n-gram ids",0);
}
rr_fread((char*) ¤t_ngram_count[i], sizeof(int),1,
temp_file[i],"temporary n-gram counts",0);
}
}
all_finished = 0;
while (!all_finished) {
/* Find the smallest current ngram */
for (i=0;i<=n-1;i++)
smallest_ngram[i] = MAX_WORDID;
for (i=0;i<=end_file-start_file;i++) {
if (!finished[i]) {
if (compare_ngrams3(smallest_ngram,current_ngram[i]) < 0) {
for (j=0;j<n;j++)
smallest_ngram[j] = current_ngram[i][j];
}
}
}
#if MAX_VOCAB_SIZE < 65535
/* This check is well-meaning but completely useless since
smallest_ngram[i] by definition cannot contain any value
greater than MAX_VOCAB_SIZE (dhuggins@cs, 2006-03) */
for (i=0;i<=n-1;i++) {
if (smallest_ngram[i] > MAX_VOCAB_SIZE) {
quit(-1,"Error : Temporary files corrupted, invalid n-gram found.\n");
}
}
#endif
/* For each of the files that are currently holding this ngram,
add its count to the temporary count, and read in a new ngram
from the files. */
temp_count = 0;
for (i=0;i<=end_file-start_file;i++) {
if (!finished[i]) {
if (compare_ngrams3(smallest_ngram,current_ngram[i]) == 0) {
temp_count = temp_count + current_ngram_count[i];
if (!rr_feof(temp_file[i])) {
for (j=0;j<=n-1;j++) {
rr_fread((char*) ¤t_ngram[i][j],sizeof(wordid_t),1,
temp_file[i],"temporary n-gram ids",0);
}
rr_fread((char*)¤t_ngram_count[i],sizeof(int),1,
temp_file[i],"temporary n-gram count",0);
}else {
finished[i] = 1;
all_finished = 1;
for (j=0;j<=end_file-start_file;j++) {
if (!finished[j])
all_finished = 0;
}
}
}
}
}
if (write_ascii) {
for (i=0;i<=n-1;i++) {
if (fprintf(outfile,"%d ",smallest_ngram[i]) < 0)
{
quit(-1,"Write error encountered while attempting to merge temporary files.\nAborting, but keeping temporary files.\n");
}
}
if (fprintf(outfile,"%d\n",temp_count) < 0)
quit(-1,"Write error encountered while attempting to merge temporary files.\nAborting, but keeping temporary files.\n");
}else {
for (i=0;i<=n-1;i++) {
rr_fwrite((char*)&smallest_ngram[i],sizeof(wordid_t),1,
outfile,"n-gram ids");
}
rr_fwrite((char*)&temp_count,sizeof(count_t),1,outfile,"n-gram counts");
}
if (fof_size > 0 && n>1) { /* Add stuff to fof arrays */
/* Code from idngram2stats */
pos_of_novelty = n;
for (i=0;i<=n-1;i++) {
if (smallest_ngram[i] > previous_ngram[i]) {
pos_of_novelty = i;
i=n;
}
}
/* Add new N-gram */
num_kgrams[n-2]++;
if (temp_count <= fof_size)
fof_array[n-2][temp_count]++;
if (!first_ngram) {
for (i=n-2;i>=MAX(1,pos_of_novelty);i--) {
num_kgrams[i-1]++;
if (ng_count[i-1] <= fof_size) {
fof_array[i-1][ng_count[i-1]]++;
}
ng_count[i-1] = temp_count;
}
}else {
for (i=n-2;i>=MAX(1,pos_of_novelty);i--) {
ng_count[i-1] = temp_count;
}
first_ngram = 0;
}
for (i=0;i<=pos_of_novelty-2;i++)
ng_count[i] += temp_count;
for (i=0;i<=n-1;i++)
previous_ngram[i]=smallest_ngram[i];
}
}
for (i=0;i<=end_file-start_file;i++) {
fclose(temp_file[i]);
remove(temp_filename[i]);
}
}
if (fof_size > 0 && n>1) { /* Display fof arrays */
/* Process last ngram */
for (i=n-2;i>=MAX(1,pos_of_novelty);i--) {
num_kgrams[i-1]++;
if (ng_count[i-1] <= fof_size)
fof_array[i-1][ng_count[i-1]]++;
ng_count[i-1] = temp_count;
}
for (i=0;i<=pos_of_novelty-2;i++)
ng_count[i] += temp_count;
display_fof_array(num_kgrams,fof_array,fof_size,stderr, n);
}
}
void merge_tempfiles (int start_file,
int end_file,
char *temp_file_root,
char *temp_file_ext,
int max_files,
FILE *outfile,
int n,
int verbosity) {
FILE *new_temp_file;
char *new_temp_filename;
FILE **temp_file;
char **temp_filename;
char **current_ngram;
char smallest_ngram[1000];
int *current_ngram_count;
flag *finished;
flag all_finished;
int temp_count;
char temp_word[500];
int i,j;
pc_message(verbosity,2,"Merging temp files %d through %d...\n", start_file,
end_file);
/*
* If we try to do more than max_files, then merge into groups,
* then merge groups recursively.
*/
if (end_file-start_file+1 > max_files) {
int new_start_file, new_end_file;
int n_file_groups = 1 + (end_file-start_file)/max_files;
fprintf(stderr, "%d files to do, in %d groups\n", end_file-start_file,
n_file_groups);
new_temp_filename = (char *) rr_malloc(300*sizeof(char));
/*
* These n_file_groups sets of files will be done in groups of
* max_files batches each, as temp files numbered
* end_file+1 ... end_file+n_file_groups,
* and then these will be merged into the final result.
*/
for (i = 0; i < n_file_groups; i++) {
/* do files i*max_files through min((i+1)*max_files-1,end_file); */
new_start_file = start_file + (i*max_files);
new_end_file = start_file + ((i+1)*max_files) - 1;
if (new_end_file > end_file) new_end_file = end_file;
sprintf(new_temp_filename,
"%s/%hu%s",
temp_file_root,
end_file+i+1,
temp_file_ext);
new_temp_file = rr_oopen(new_temp_filename);
merge_tempfiles(new_start_file,
new_end_file,
temp_file_root,
temp_file_ext,
max_files,
new_temp_file,
n,
verbosity);
rr_iclose(new_temp_file);
}
merge_tempfiles(end_file+1,
end_file+n_file_groups,
temp_file_root,
temp_file_ext,
max_files,
outfile,
n,
verbosity);
return;
}
/*
* We know we are now doing <= max_files.
*/
temp_file = (FILE **) rr_malloc((end_file+1)*sizeof(FILE *));
temp_filename = (char **) rr_malloc((end_file+1)*sizeof(char *));
for (i=start_file;i<=end_file;i++) {
temp_filename[i] = (char *) rr_malloc(300*sizeof(char));
}
current_ngram = (char **) rr_malloc((end_file+1)*sizeof(char *));
for (i=start_file;i<=end_file;i++) {
current_ngram[i] = (char *) rr_malloc(1000*sizeof(char));
}
current_ngram_count = (int *) rr_malloc((end_file+1)*sizeof(int));
finished = (flag *) rr_malloc(sizeof(flag)*(end_file+1));
/* Open all the temp files for reading */
for (i=start_file;i<=end_file;i++) {
sprintf(temp_filename[i],"%s/%hu%s",
temp_file_root,i,temp_file_ext);
temp_file[i] = rr_iopen(temp_filename[i]);
}
/* Now go through the files simultaneously, and write out the appropriate
ngram counts to the output file. */
for (i=start_file;i<=end_file;i++) {
finished[i] = 0;
if (!rr_feof(temp_file[i])) {
for (j=0;j<=n-1;j++) {
if (fscanf(temp_file[i],"%s",temp_word) != 1) {
if (!rr_feof(temp_file[i]))
quit(-1,"Error reading temp file %s\n",temp_filename[i]);
}else {
if (j==0)
strcpy(current_ngram[i],temp_word);
else {
strcat(current_ngram[i]," ");
strcat(current_ngram[i],temp_word);
}
}
}
if (fscanf(temp_file[i],"%d",¤t_ngram_count[i]) != 1) {
if (!rr_feof(temp_file[i]))
quit(-1,"Error reading temp file %s\n",temp_filename[i]);
}
}
}
all_finished = 0;
while (!all_finished) {
/* Find the smallest current ngram */
strcpy(smallest_ngram,"");
for (i=start_file;i<=end_file;i++) {
if (!finished[i]) {
if (strcmp(smallest_ngram,current_ngram[i]) > 0 ||
(smallest_ngram[0] == '\0'))
strcpy(smallest_ngram,current_ngram[i]);
}
}
/* For each of the files that are currently holding this ngram,
add its count to the temporary count, and read in a new ngram
from the files. */
temp_count = 0;
for (i=start_file;i<=end_file;i++) {
if (!finished[i]) {
if (!strcmp(smallest_ngram,current_ngram[i])) {
temp_count += current_ngram_count[i];
if (!rr_feof(temp_file[i])) {
for (j=0;j<=n-1;j++) {
if (fscanf(temp_file[i],"%s",temp_word) != 1) {
if (!rr_feof(temp_file[i])) {
quit(-1,"Error reading temp file %s\n",temp_filename[i]);
}
}else {
if (j==0)
strcpy(current_ngram[i],temp_word);
else {
strcat(current_ngram[i]," ");
strcat(current_ngram[i],temp_word);
}
}
}
if (fscanf(temp_file[i],"%d",¤t_ngram_count[i]) != 1) {
if (!rr_feof(temp_file[i])) {
quit(-1,"Error reading temp file count %s\n",
temp_filename[i]);
}
}
}
/*
* PWP: Note that the fscanf may have changed the state of
* temp_file[i], so we re-ask the question rather than just
* doing an "else".
*/
if (rr_feof(temp_file[i])) {
finished[i] = 1;
all_finished = 1;
for (j=start_file;j<=end_file;j++) {
if (!finished[j]) {
all_finished = 0;
}
}
}
}
}
}
/*
* PWP: We cannot conditionalize this on (!all_finished) because
* if we do we may have lost the very last count. (Consider the
* case when several files have ran out of data, but the last
* couple have the last count in them.)
*/
if (fprintf(outfile,"%s %d\n",smallest_ngram,temp_count) < 0) {
quit(-1,"Write error encountered while attempting to merge temporary files.\nAborting, but keeping temporary files.\n");
}
}
for (i=start_file;i<=end_file;i++) {
rr_iclose(temp_file[i]);
remove(temp_filename[i]);
}
free(temp_file);
for (i=start_file;i<=end_file;i++) {
free(temp_filename[i]);
}
free(temp_filename);
for (i=start_file;i<=end_file;i++) {
free(current_ngram[i]);
}
free(current_ngram);
free(current_ngram_count);
free(finished);
}
/*
@return number_of_tempfiles
*/
int read_txt2ngram_buffer(FILE* infp,
struct idngram_hash_table *vocabulary,
int32 verbosity,
wordid_t *buffer,
int buffer_size,
unsigned int n,
char* temp_file_root,
char* temp_file_ext,
FILE* temp_file
)
{
/* Read text into buffer */
char temp_word[MAX_WORD_LENGTH];
int position_in_buffer;
int number_of_tempfiles;
unsigned int i,j;
wordid_t *placeholder;
wordid_t *temp_ngram;
int temp_count;
#if 1
int tmpval;
#endif
temp_ngram = (wordid_t *) rr_malloc(sizeof(wordid_t)*n);
placeholder = (wordid_t *) rr_malloc(sizeof(wordid_t)*n);
ng=n;
position_in_buffer = 0;
number_of_tempfiles = 0;
//tk: looks like things may croak if the corpus has less than n words
//not that such a corpus would be useful anyway
for (i=0;i<=n-1;i++) {
get_word(infp,temp_word);
/*
fprintf(stderr,"%s \n",temp_word);
fprintf(stderr,"%d \n",index2(vocabulary,temp_word));
fflush(stderr);
*/
add_to_buffer(index2(vocabulary,temp_word),0,i,buffer);
}
while (!rr_feof(infp)) {
/* Fill up the buffer */
pc_message(verbosity,2,"Reading text into the n-gram buffer...\n");
pc_message(verbosity,2,"20,000 n-grams processed for each \".\", 1,000,000 for each line.\n");
while ((position_in_buffer<buffer_size) && (!rr_feof(infp))) {
position_in_buffer++;
show_idngram_nlines(position_in_buffer,verbosity);
for (i=1;i<=n-1;i++)
add_to_buffer(buffer_contents(position_in_buffer-1,i,buffer),
position_in_buffer,i-1,buffer);
if (get_word(infp,temp_word) == 1) {
/*
fprintf(stderr,"%s \n",temp_word);
fprintf(stderr,"%d \n",index2(vocabulary,temp_word));
fflush(stderr);
*/
add_to_buffer(index2(vocabulary,temp_word),position_in_buffer,
n-1,buffer);
}
}
for (i=0;i<=n-1;i++)
placeholder[i] = buffer_contents(position_in_buffer,i,buffer);
/* Sort buffer */
pc_message(verbosity,2,"\nSorting n-grams...\n");
qsort((void*) buffer,(size_t) position_in_buffer,n*sizeof(wordid_t),compare_ngrams);
/* Output the buffer to temporary BINARY file */
number_of_tempfiles++;
sprintf(temp_word,"%s/%hu%s",temp_file_root,
number_of_tempfiles,temp_file_ext);
pc_message(verbosity,2,"Writing sorted n-grams to temporary file %s\n",
temp_word);
temp_file = rr_oopen(temp_word);
for (i=0;i<=n-1;i++) {
temp_ngram[i] = buffer_contents(0,i,buffer);
#if MAX_VOCAB_SIZE < 65535
/* This check is well-meaning but completely useless since
buffer_contents() can never return something greater than
MAX_VOCAB_SIZE (dhuggins@cs, 2006-03) */
if (temp_ngram[i] > MAX_VOCAB_SIZE)
quit(-1,"Invalid trigram in buffer.\nAborting");
#endif
}
temp_count = 1;
for (i=1;i<=position_in_buffer;i++) {
tmpval=compare_ngrams(temp_ngram,&buffer[i*n]);
/* for(k=0;k<=n-1;k++){
fprintf(stderr, "tmpval: %d k %d, temp_ngram %d, &buffer[i*n] %d\n",tmpval, k, temp_ngram[k], (&buffer[i*n])[k]);
}*/
if (!compare_ngrams(temp_ngram,&buffer[i*n]))
temp_count++;
else {
/* printf("Have been here?\n");*/
for (j=0;j<=n-1;j++) {
rr_fwrite((char*) &temp_ngram[j],sizeof(wordid_t),1,
temp_file,"temporary n-gram ids");
temp_ngram[j] = buffer_contents(i,j,buffer);
}
rr_fwrite((char*)&temp_count,sizeof(int),1,temp_file,
"temporary n-gram counts");
/* for(j=0 ; j<=n-1;j++)
fprintf(stderr,"%d ",temp_ngram[j]);
fprintf(stderr,"%d\n",temp_count);*/
temp_count = 1;
}
}
rr_oclose(temp_file);
for (i=0;i<=n-1;i++)
add_to_buffer(placeholder[i],0,i,buffer);
position_in_buffer = 0;
}
return number_of_tempfiles;
}
/* To make this function less dependent on input stream, just pull records out and create an interface for it
*/
int wfreq2vocab_impl(FILE* ifp, FILE* ofp, int cutoff, int vocab_size, int num_recs, int verbosity)
{
flag gt_set;
flag top_set;
int current_rec;
int num_above_threshold;
int num_to_output;
int i;
word_rec *records;
char temp_word[750];
gt_set = (cutoff != -1);
top_set = (vocab_size != -1);
if(cutoff==-1) cutoff=0;
if(vocab_size==-1) vocab_size=0;
if (gt_set && top_set)
quit(-1,"wfreq2vocab : Error : Can't use both the -top and the -gt options.\n");
if (!gt_set && !top_set)
vocab_size = 20000;
if (gt_set)
pc_message(verbosity,2,"wfreq2vocab : Will generate a vocabulary containing all words which\n occurred more that %d times. Reading wfreq stream from stdin...\n",cutoff);
else
pc_message(verbosity,2,"wfreq2vocab : Will generate a vocabulary containing the most\n frequent %d words. Reading wfreq stream from stdin...\n",vocab_size);
current_rec = 0;
num_above_threshold = 0;
records = (word_rec *) rr_malloc(sizeof(word_rec)*num_recs);
while (!rr_feof(ifp)) {
if (fscanf(ifp, "%s %d",temp_word,&(records[current_rec].count)) != 2) {
if (!rr_feof(ifp))
quit(-1,"Error reading unigram counts from standard input.\n");
}else {
records[current_rec].word = salloc(temp_word);
if (gt_set && records[current_rec].count > cutoff)
num_above_threshold++;
current_rec++;
}
if(current_rec > num_recs ){
quit2(-1,"The number of records %d reach the user-defined limit %d, consider to increase the number of records by -records\n",current_rec,num_recs);
}
}
/* Sort records in descending order of count */
qsort((void*) records,(size_t) current_rec, sizeof(word_rec),sort_by_count);
if (gt_set)
num_to_output = num_above_threshold;
else
num_to_output = vocab_size;
if (current_rec<num_to_output)
num_to_output = current_rec;
/* Now sort the relevant records alphabetically */
qsort((void*) records,(size_t) num_to_output, sizeof(word_rec),sort_alpha);
if (gt_set)
pc_message(verbosity,2,"Size of vocabulary = %d\n",num_to_output);
if (num_to_output>MAX_UNIGRAM) {
pc_message(verbosity,1,"Warning : Vocab size exceeds %d. This might cause problems with \n",MAX_UNIGRAM);
pc_message(verbosity,1,"other tools, since word id's are stored in 2 bytes.\n");
}
if (num_to_output == 0)
pc_message(verbosity,1,"Warning : Vocab size = 0.\n");
/* Print the vocab to stdout */
printf("## Vocab generated by v2 of the CMU-Cambridge Statistcal\n");
printf("## Language Modeling toolkit.\n");
printf("##\n");
printf("## Includes %d words ",num_to_output);
printf("##\n");
for (i=0;i<=num_to_output-1;i++)
fprintf(ofp,"%s\n",records[i].word);
pc_message(verbosity,0,"wfreq2vocab : Done.\n");
return 0;
}
int main(int argc, char **argv) {
int i,j;
ng_t* ng;
int verbosity;
int mem_alloc_method; /* Method used to decide how much memory to
allocate for count tables */
int buffer_size;
flag is_ascii;
ngram current_ngram;
ngram previous_ngram;
count_t *ng_count; /* Array indicating the number of occurrances of
the current 1-gram, 2-gram, ... ,n-gram
Size depends on #define in general.h
*/
int nlines;
int pos_of_novelty;
int prev_id1;
flag contains_unks;
int mem_alloced;
flag displayed_oov_warning; /** Display OOV warning
*/
/* ------------------ Process command line --------------------- */
report_version(&argc,argv);
if (argc == 1 || pc_flagarg(&argc, argv,"-help")) {
/* Display help message */
help_message();
exit(1);
}
verbosity = pc_intarg(&argc, argv,"-verbosity",DEFAULT_VERBOSITY);
/* Initialization */
{
ng=init_ng(
&argc,
argv,
verbosity
);
mem_alloc_method = init_alloc_method(ng, &argc, argv, &buffer_size);
if (!strcmp(ng->id_gram_filename,"-") && mem_alloc_method == TWO_PASSES)
quit(-1,"Error: If idngram is read from stdin, then cannot use -calc_mem option.\n");
is_ascii = set_lmformat(pc_flagarg(&argc,argv,"-ascii_input"),
pc_flagarg(&argc,argv,"-bin_input"),
ng);
/* Report parameters */
report_param(verbosity,ng,
is_ascii, mem_alloc_method, buffer_size);
pc_report_unk_args(&argc,argv,verbosity);
}
/* --------------- Read in the vocabulary -------------- */
read_vocab(ng,verbosity);
/* --------------- Allocate space for the table_size array --------- */
init_ng_table_size(ng,
mem_alloc_method,
is_ascii,
verbosity,
buffer_size
);
/* ----------- Allocate memory for tree structure -------------- */
ng->count = NULL;
ng->count4 = NULL;
ng->marg_counts = NULL;
ng->marg_counts4 = NULL;
ng->count_table = NULL;
ng->count = (count_ind_t **) rr_malloc(sizeof(count_ind_t *)*ng->n);
ng->count4 = (count_t **) rr_malloc(sizeof(count_t *)*ng->n);
ng->count_table = (count_t **) rr_malloc(sizeof(count_t *)*ng->n);
if (ng->four_byte_counts) {
ng->marg_counts4 = (count_t *) rr_calloc(sizeof(count_t), ng->table_sizes[0]);
}else {
for (i=0;i<=ng->n-1;i++)
ng->count_table[i] = (count_t *) rr_calloc(ng->count_table_size+1,
sizeof(count_t));
ng->marg_counts = (count_ind_t *) rr_calloc(sizeof(count_ind_t),ng->table_sizes[0]);
fprintf(stderr, "table_size %d\n",ng->table_sizes[0]);
fflush(stderr);
}
ng->word_id = (id__t **) rr_malloc(sizeof(id__t *)*ng->n);
if (ng->four_byte_alphas) {
ng->bo_weight4 = (four_byte_t **) rr_malloc(sizeof(four_byte_t *)*ng->n);
ng->bo_weight4[0] = (four_byte_t *) rr_malloc(sizeof(four_byte_t)*
ng->table_sizes[0]);
}else {
ng->bo_weight = (bo_weight_t **) rr_malloc(sizeof(bo_weight_t *)*ng->n);
ng->bo_weight[0] = (bo_weight_t *) rr_malloc(sizeof(bo_weight_t)*
ng->table_sizes[0]);
}
ng->ind = (index__t **) rr_malloc(sizeof(index__t *)*ng->n);
/* First table */
if (ng->four_byte_counts)
ng->count4[0] = (count_t *) rr_calloc(ng->table_sizes[0],sizeof(count_t));
else
ng->count[0] = (count_ind_t *) rr_calloc(ng->table_sizes[0],sizeof(count_ind_t));
ng->uni_probs = (uni_probs_t *) rr_malloc(sizeof(uni_probs_t)*
ng->table_sizes[0]);
ng->uni_log_probs = (uni_probs_t *) rr_malloc(sizeof(uni_probs_t)*
ng->table_sizes[0]);
if (ng->n >=2)
ng->ind[0] = (index__t *) rr_calloc(ng->table_sizes[0],sizeof(index__t));
for (i=1;i<=ng->n-2;i++) {
ng->word_id[i] = (id__t *) rr_malloc(sizeof(id__t)*ng->table_sizes[i]);
if (ng->four_byte_counts)
ng->count4[i] = (count_t *) rr_malloc(sizeof(count_t)*ng->table_sizes[i]);
else
ng->count[i] = (count_ind_t *) rr_malloc(sizeof(count_ind_t)*ng->table_sizes[i]);
if (ng->four_byte_alphas)
ng->bo_weight4[i] = (four_byte_t *) rr_malloc(sizeof(four_byte_t)*ng->table_sizes[i]);
else
ng->bo_weight[i] = (bo_weight_t *) rr_malloc(sizeof(bo_weight_t)*ng->table_sizes[i]);
ng->ind[i] = (index__t *) rr_malloc(sizeof(index__t)*ng->table_sizes[i]);
mem_alloced = sizeof(count_ind_t) + sizeof(bo_weight_t) +
sizeof(index__t) + sizeof(id__t);
if (ng->four_byte_alphas)
mem_alloced += 4;
mem_alloced *= ng->table_sizes[i];
pc_message(verbosity,2,"Allocated %d bytes to table for %d-grams.\n",
mem_alloced,i+1);
}
ng->word_id[ng->n-1] = (id__t *)
rr_malloc(sizeof(id__t)*ng->table_sizes[ng->n-1]);
if (ng->four_byte_counts)
ng->count4[ng->n-1] = (count_t *) rr_malloc(sizeof(count_t)*ng->table_sizes[ng->n-1]);
else
ng->count[ng->n-1] = (count_ind_t *) rr_malloc(sizeof(count_ind_t)*ng->table_sizes[ng->n-1]);
pc_message(verbosity,2,"Allocated (%d+%d) bytes to table for %d-grams.\n",
ng->four_byte_counts?sizeof(count_t):sizeof(count_ind_t),
sizeof(id__t)*ng->table_sizes[ng->n-1],ng->n);
/* Allocate memory for table for first-byte of indices */
ng_allocate_ptr_table(ng,NULL,0);
/* Allocate memory for alpha array */
ng->alpha_array = (double *) rr_malloc(sizeof(double)*ng->out_of_range_alphas);
ng->size_of_alpha_array = 0;
/* Allocate memory for frequency of frequency information */
ng->freq_of_freq = (fof_t **) rr_malloc(sizeof(fof_t *)*ng->n);
NG_DISC_METH(ng)->allocate_freq_of_freq(ng);
/* Read n-grams into the tree */
pc_message(verbosity,2,"Processing id n-gram file.\n");
pc_message(verbosity,2,"20,000 n-grams processed for each \".\", 1,000,000 for each line.\n");
/* Allocate space for ngrams id arrays */
current_ngram.id_array = (id__t *) rr_calloc(ng->n,sizeof(id__t));
previous_ngram.id_array = (id__t *) rr_calloc(ng->n,sizeof(id__t));
current_ngram.n = ng->n;
previous_ngram.n = ng->n;
ng->num_kgrams = (ngram_sz_t *) rr_calloc(ng->n,sizeof(ngram_sz_t));
ng_count = (count_t *) rr_calloc(ng->n,sizeof(count_t));
nlines = 1;
ng->n_unigrams = 0;
/* Process first n-gram */
get_ngram(ng->id_gram_fp,¤t_ngram,is_ascii);
contains_unks = ngram_chk_contains_unks(¤t_ngram,ng->n);
/* Skip over any unknown words. They will come first, because <UNK>
always has a word ID of zero. */
while (ng->vocab_type == CLOSED_VOCAB && contains_unks){
/* Stop looking if there are no more N-Grams. Of course, this
means training will fail, since there are no unigrams. */
if (get_ngram(ng->id_gram_fp,¤t_ngram,is_ascii) == 0)
break;
contains_unks = ngram_chk_contains_unks(¤t_ngram,ng->n);
}
for (i=0;i<=ng->n-2;i++) {
ng->ind[i][0] = new_index(0,ng->ptr_table[i],&(ng->ptr_table_size[i]),0);
ng->word_id[i+1][0] = current_ngram.id_array[i+1];
ng->num_kgrams[i+1]++;
ng_count[i] = current_ngram.count;
}
ng_count[0] = current_ngram.count;
NG_DISC_METH(ng)->update_freq_of_freq(ng,ng->n-1,current_ngram.count);
store_normal_count(ng,0,current_ngram.count,ng->n-1);
if (current_ngram.count <= ng->cutoffs[ng->n-2])
ng->num_kgrams[ng->n-1]--;
ngram_copy(&previous_ngram,¤t_ngram,ng->n);
prev_id1 = current_ngram.id_array[0];
displayed_oov_warning = 0;
while (!rr_feof(ng->id_gram_fp)) {
if (get_ngram(ng->id_gram_fp,¤t_ngram,is_ascii)) {
if (ng->vocab_type == CLOSED_VOCAB)
contains_unks=ngram_chk_contains_unks(¤t_ngram,ng->n);
if (!contains_unks || ng->vocab_type != CLOSED_VOCAB) {
/* Test for where this ngram differs from last - do we have an
out-of-order ngram? */
pos_of_novelty = ngram_find_pos_of_novelty(¤t_ngram,&previous_ngram,ng->n,nlines);
nlines++;
show_idngram_nlines(nlines, verbosity);
/* Add new n-gram as soon as it is encountered */
/* If all of the positions 2,3,...,n of the n-gram are context
cues then ignore the n-gram. */
if (ng->n > 1) {
NG_DISC_METH(ng)->update_freq_of_freq(ng,ng->n-1,current_ngram.count);
store_normal_count(ng,ng->num_kgrams[ng->n-1],current_ngram.count,ng->n-1);
ng->word_id[ng->n-1][ng->num_kgrams[ng->n-1]] = current_ngram.id_array[ng->n-1];
ng->num_kgrams[ng->n-1]++;
if (ng->num_kgrams[ng->n-1] >= ng->table_sizes[ng->n-1])
quit(-1,"\nMore than %d %d-grams needed to be stored. Rerun with a higher table size.\n",ng->table_sizes[ng->n-1],ng->n);
}
/* Deal with new 2,3,...,(n-1)-grams */
for (i=ng->n-2;i>=MAX(1,pos_of_novelty);i--) {
NG_DISC_METH(ng)->update_freq_of_freq(ng,i,ng_count[i]);
if (ng_count[i] <= ng->cutoffs[i-1])
ng->num_kgrams[i]--;
else
store_normal_count(ng,ng->num_kgrams[i]-1,ng_count[i],i);
ng_count[i] = current_ngram.count;
ng->word_id[i][ng->num_kgrams[i]] = current_ngram.id_array[i];
ng->ind[i][ng->num_kgrams[i]] = new_index(ng->num_kgrams[i+1]-1,
ng->ptr_table[i],
&(ng->ptr_table_size[i]),
ng->num_kgrams[i]);
ng->num_kgrams[i]++;
if (ng->num_kgrams[i] >= ng->table_sizes[i])
quit(-1,"More than %d %d-grams needed to be stored. Rerun with a higher table size.\n",ng->table_sizes[i],i+1);
}
for (i=0;i<=pos_of_novelty-1;i++)
ng_count[i] += current_ngram.count;
/* Deal with new 1-grams */
if (pos_of_novelty == 0) {
if (ng->n>1) {
for (i = prev_id1 + 1; i <= current_ngram.id_array[0]; i++) {
ng->ind[0][i] = new_index(ng->num_kgrams[1]-1,
ng->ptr_table[0],
&(ng->ptr_table_size[0]),
i);
}
prev_id1 = current_ngram.id_array[0];
}
NG_DISC_METH(ng)->update_freq_of_freq(ng,0,ng_count[0]);
if (!ng->context_cue[previous_ngram.id_array[0]]) {
ng->n_unigrams += ng_count[0];
store_normal_count(ng,previous_ngram.id_array[0],ng_count[0],0);
}
store_marginal_count(ng,previous_ngram.id_array[0],ng_count[0],0);
ng_count[0] = current_ngram.count;
}
if (current_ngram.count <= ng->cutoffs[ng->n-2])
ng->num_kgrams[ng->n-1]--;
ngram_copy(&previous_ngram,¤t_ngram,ng->n);
}else {
if (!displayed_oov_warning){
pc_message(verbosity,2,"Warning : id n-gram stream contains OOV's (n-grams will be ignored).\n");
displayed_oov_warning = 1;
}
}
}
}
rr_iclose(ng->id_gram_fp);
for (i=ng->n-2;i>=1;i--) {
NG_DISC_METH(ng)->update_freq_of_freq(ng,i,ng_count[i]);
if (ng_count[i] <= ng->cutoffs[i-1])
ng->num_kgrams[i]--;
else
store_normal_count(ng,ng->num_kgrams[i]-1,ng_count[i],i);
}
NG_DISC_METH(ng)->update_freq_of_freq(ng,0,ng_count[0]);
if (!ng->context_cue[current_ngram.id_array[0]]) {
ng->n_unigrams += ng_count[0];
store_normal_count(ng,current_ngram.id_array[0],ng_count[0],0);
}
store_marginal_count(ng,current_ngram.id_array[0],ng_count[0],0);
if (ng->n>1) {
for (i=current_ngram.id_array[0]+1;i<=ng->vocab_size;i++)
ng->ind[0][i] = new_index(ng->num_kgrams[1],
ng->ptr_table[0],
&(ng->ptr_table_size[0]),
current_ngram.id_array[0]);
}
/* The idngram reading is completed at this point */
pc_message(verbosity,2,"\n");
/* Impose a minimum unigram count, if required */
if (ng->min_unicount > 0) {
int nchanged= 0;
for (i=ng->first_id;i<=ng->vocab_size;i++) {
if ((return_count(ng->four_byte_counts,
ng->count_table[0],
ng->count[0],
ng->count4[0],
i) < ng->min_unicount) && !ng->context_cue[i]) {
/* There was a bug in V2's switch. Look at segment for ABSOLUTE */
NG_DISC_METH(ng)->reduce_ug_freq_of_freq(ng,i);
ng->n_unigrams += (ng->min_unicount - ng->count[0][i]);
store_normal_count(ng,i,ng->min_unicount,0);
nchanged++;
}
}
if (nchanged > 0)
pc_message(verbosity,2,
"Unigram counts of %d words were bumped up to %d.\n",
nchanged,ng->min_unicount);
}
/* Count zeroton information for unigrams */
ng->freq_of_freq[0][0] = 0;
for (i=ng->first_id;i<=ng->vocab_size;i++) {
if (return_count(ng->four_byte_counts,
ng->count_table[0],
ng->count[0],
ng->count4[0],
i) == 0) {
ng->freq_of_freq[0][0]++;
}
}
if (ng->discounting_method == GOOD_TURING) {
for (i=0;i<=ng->n-1;i++)
for (j=1;j<=ng->fof_size[i];j++)
pc_message(verbosity,3,"fof[%d][%d] = %d\n",i,j,ng->freq_of_freq[i][j]);
}
pc_message(verbosity,2,"Calculating discounted counts.\n");
NG_DISC_METH(ng)->compute_discount_aux(ng, verbosity);
/* Smooth unigram distribution, to give some mass to zerotons */
compute_unigram(ng,verbosity);
/* Increment Contexts if using Good-Turing discounting-> No need otherwise,
since all values are discounted anyway. */
if (ng->discounting_method == GOOD_TURING) {
pc_message(verbosity,2,"Incrementing contexts...\n");
for (i=ng->n-1;i>=1;i--)
increment_context(ng,i,verbosity);
}
/* Calculate back-off weights */
pc_message(verbosity,2,"Calculating back-off weights...\n");
for (i=1;i<=ng->n-1;i++)
compute_back_off(ng,i,verbosity);
if (!ng->four_byte_alphas)
pc_message(verbosity,3,"Number of out of range alphas = %d\n",
ng->size_of_alpha_array);
/* Write out LM */
pc_message(verbosity,2,"Writing out language model...\n");
if (ng->write_arpa)
write_arpa_lm(ng,verbosity);
if (ng->write_bin)
write_bin_lm(ng,verbosity);
pc_message(verbosity,0,"idngram2lm : Done.\n");
return 0;
}
