rank_bm25(cache_config& cconfig) {
uint64_t num_terms;
load_from_cache(num_terms, surf::KEY_COLLEN, cconfig);
if (!cache_file_exists(surf::KEY_DOC_LENGTHS, cconfig)){
surf::construct_doc_lengths<sdsl::int_alphabet_tag::WIDTH>(cconfig);
}
load_from_cache(doc_lengths, surf::KEY_DOC_LENGTHS, cconfig);
num_docs = doc_lengths.size();
std::cerr<<"num_docs = "<<num_docs<<std::endl;
avg_doc_len = (double)num_terms / (double)num_docs;
std::cerr<<"avg_doc_len = "<<avg_doc_len<<std::endl;
}
rank_tfidf(cache_config& cconfig) {
load_from_cache(num_terms, surf::KEY_COLLEN, cconfig);
if (!cache_file_exists(surf::KEY_DOC_LENGTHS, cconfig)){
surf::construct_doc_lengths<sdsl::int_alphabet_tag::WIDTH>(cconfig);
}
load_from_cache(doc_lengths, surf::KEY_DOC_LENGTHS, cconfig);
num_docs = doc_lengths.size();
std::cerr<<"num_docs = "<<num_docs<<std::endl;
auto min_itr = std::min_element(doc_lengths.begin(),doc_lengths.end());
min_doc_len = *min_itr;
std::cerr<<"min_doc_len = "<<min_doc_len<<std::endl;
}
void construct_lcp_PHI(cache_config& config)
{
static_assert(t_width == 0 or t_width == 8 , "construct_lcp_PHI: width must be `0` for integer alphabet and `8` for byte alphabet");
typedef int_vector<>::size_type size_type;
typedef int_vector<t_width> text_type;
const char* KEY_TEXT = key_text_trait<t_width>::KEY_TEXT;
int_vector_buffer<> sa_buf(config.file_map[conf::KEY_SA]);
size_type n = sa_buf.size();
assert(n > 0);
if (1 == n) { // Handle special case: Input only the sentinel character.
int_vector<> lcp(1, 0);
store_to_cache(lcp, conf::KEY_LCP, config);
return;
}
// (1) Calculate PHI (stored in array plcp)
int_vector<> plcp(n, 0, sa_buf.width());
for (size_type i=0, sai_1 = 0; i < n; ++i) {
size_type sai = sa_buf[i];
plcp[ sai ] = sai_1;
sai_1 = sai;
}
// (2) Load text from disk
text_type text;
load_from_cache(text, KEY_TEXT, config);
// (3) Calculate permuted LCP array (text order), called PLCP
size_type max_l = 0;
for (size_type i=0, l=0; i < n-1; ++i) {
size_type phii = plcp[i];
while (text[i+l] == text[phii+l]) {
++l;
}
plcp[i] = l;
if (l) {
max_l = std::max(max_l, l);
--l;
}
}
util::clear(text);
uint8_t lcp_width = bits::hi(max_l)+1;
// (4) Transform PLCP into LCP
std::string lcp_file = cache_file_name(conf::KEY_LCP, config);
size_type buffer_size = 1000000; // buffer_size is a multiple of 8!
int_vector_buffer<> lcp_buf(lcp_file, std::ios::out, buffer_size, lcp_width); // open buffer for lcp
lcp_buf[0] = 0;
sa_buf.buffersize(buffer_size);
for (size_type i=1; i < n; ++i) {
size_type sai = sa_buf[i];
lcp_buf[i] = plcp[sai];
}
lcp_buf.close();
register_cache_file(conf::KEY_LCP, config);
}
void construct_lcp_kasai(cache_config& config)
{
int_vector<> lcp;
typedef int_vector<>::size_type size_type;
construct_isa(config);
{
int_vector<t_width> text;
if (!load_from_cache(text, key_text_trait<t_width>::KEY_TEXT, config)) {
return;
}
int_vector_file_buffer<> isa_buf(config.file_map[constants::KEY_ISA], 1000000); // init isa file_buffer
int_vector<> sa;
if (!load_from_cache(sa, constants::KEY_SA, config)) {
return;
}
// use Kasai algorithm to compute the lcp values
for (size_type i=0,j=0,sa_1=0,l=0, r_sum=0, r=isa_buf.load_next_block(), n=isa_buf.int_vector_size; r_sum < n;) {
for (; i < r_sum+r; ++i) {
sa_1 = isa_buf[i-r_sum]; // = isa[i]
if (sa_1) {
j = sa[sa_1-1];
if (l) --l;
assert(i!=j);
while (text[i+l]==text[j+l]) { // i+l < n and j+l < n are not necessary, since text[n]=0 and text[i]!=0 (i<n) and i!=j
++l;
}
sa[ sa_1-1 ] = l; //overwrite sa array with lcp values
} else {
l = 0;
sa[ n-1 ] = 0;
}
}
r_sum += r;
r = isa_buf.load_next_block();
}
for (size_type i=sa.size(); i>1; --i) {
sa[i-1] = sa[i-2];
}
sa[0] = 0;
lcp.swap(sa);
}
store_to_cache(lcp, constants::KEY_LCP, config);
}
void construct_lcp_kasai(cache_config& config)
{
static_assert(t_width == 0 or t_width == 8 , "construct_lcp_kasai: width must be `0` for integer alphabet and `8` for byte alphabet");
int_vector<> lcp;
typedef int_vector<>::size_type size_type;
construct_isa(config);
{
int_vector<t_width> text;
if (!load_from_cache(text, key_text_trait<t_width>::KEY_TEXT, config)) {
return;
}
int_vector_buffer<> isa_buf(config.file_map[conf::KEY_ISA], std::ios::in, 1000000); // init isa file_buffer
int_vector<> sa;
if (!load_from_cache(sa, conf::KEY_SA, config)) {
return;
}
// use Kasai algorithm to compute the lcp values
for (size_type i=0,j=0,sa_1=0,l=0, n=isa_buf.size(); i < n; ++i) {
sa_1 = isa_buf[i]; // = isa[i]
if (sa_1) {
j = sa[sa_1-1];
if (l) --l;
assert(i!=j);
while (text[i+l]==text[j+l]) { // i+l < n and j+l < n are not necessary, since text[n]=0 and text[i]!=0 (i<n) and i!=j
++l;
}
sa[ sa_1-1 ] = l; //overwrite sa array with lcp values
} else {
l = 0;
sa[ n-1 ] = 0;
}
}
for (size_type i=sa.size(); i>1; --i) {
sa[i-1] = sa[i-2];
}
sa[0] = 0;
lcp.swap(sa);
}
store_to_cache(lcp, conf::KEY_LCP, config);
}
std::string cache_read() {
PROFILE_FUNC();
std::string data;
PROFILE_START(action_find); // Starts new action which will be inner to ACTION_READ
bool found = find_record();
PROFILE_STOP(action_find);
if (!found) {
PROFILE_BLOCK(load_from_disk);
data = read_from_disk();
put_into_cache(data);
return data; // Here all action guards are destructed and actions are correctly finished
}
data = load_from_cache();
return data;
}
int main( int argc, char** argv ) {
/* parse command line */
cmdargs_t args = parse_args(argc,argv);
/* parse repo */
auto cc = surf::parse_collection(args.collection_dir);
sdsl::int_vector_buffer<> T(args.collection_dir+"/"+surf::TEXT_FILENAME);
std::cout << "n = |T|= " << T.size() << std::endl;
surf::construct_doc_cnt<sdsl::int_alphabet_tag::WIDTH>(cc);
uint64_t doc_cnt = 0;
load_from_cache(doc_cnt, surf::KEY_DOCCNT, cc);
std::cout << "number of documents = N = " << doc_cnt << std::endl;
std::ifstream dic_fs(args.collection_dir+"/"+surf::DICT_FILENAME);
std::string line;
size_t num_terms = 0;
while( std::getline(dic_fs,line) ) {
num_terms++;
}
std::cout << "number of terms = sigma = " << num_terms << std::endl;
std::cout << "avg document length = " << T.size() / doc_cnt << std::endl;
}
void construct_sa(cache_config& config)
{
static_assert(t_width == 0 or t_width == 8 , "construct_sa: width must be `0` for integer alphabet and `8` for byte alphabet");
const char* KEY_TEXT = key_text_trait<t_width>::KEY_TEXT;
if (t_width == 8) {
typedef int_vector<t_width> text_type;
text_type text;
load_from_cache(text, KEY_TEXT, config);
// call divsufsort
int_vector<> sa(text.size(), 0, bits::hi(text.size())+1);
algorithm::calculate_sa((const unsigned char*)text.data(), text.size(), sa);
store_to_cache(sa, conf::KEY_SA, config);
} else if (t_width == 0) {
// call qsufsort
int_vector<> sa;
sdsl::qsufsort::construct_sa(sa, config.file_map[KEY_TEXT].c_str(), 0);
store_to_cache(sa, conf::KEY_SA, config);
} else {
std::cerr << "Unknown alphabet type" << std::endl;
}
}
void construct_lcp_PHI(cache_config& config)
{
typedef int_vector<>::size_type size_type;
typedef int_vector<t_width> text_type;
const char* KEY_TEXT = key_text_trait<t_width>::KEY_TEXT;
int_vector_file_buffer<> sa_buf(config.file_map[constants::KEY_SA]);
size_type n = sa_buf.int_vector_size;
assert(n > 0);
if (1 == n) { // Handle special case: Input only the sentinel character.
int_vector<> lcp(1, 0);
store_to_cache(lcp, constants::KEY_LCP, config);
return;
}
// (1) Calculate PHI (stored in array plcp)
int_vector<> plcp(n, 0, sa_buf.width);
for (size_type i=0, r_sum=0, r=sa_buf.load_next_block(), sai_1 = 0; r_sum < n;) {
for (; i < r_sum+r; ++i) {
size_type sai = sa_buf[i-r_sum];
plcp[ sai ] = sai_1;
sai_1 = sai;
}
r_sum += r; r = sa_buf.load_next_block();
}
// (2) Load text from disk
text_type text;
load_from_cache(text, KEY_TEXT, config);
// (3) Calculate permuted LCP array (text order), called PLCP
size_type max_l = 0;
for (size_type i=0, l=0; i < n-1; ++i) {
size_type phii = plcp[i];
while (text[i+l] == text[phii+l]) {
++l;
}
plcp[i] = l;
if (l) {
max_l = std::max(max_l, l);
--l;
}
}
util::clear(text);
uint8_t lcp_width = bits::hi(max_l)+1;
// (4) Transform PLCP into LCP
std::string lcp_file = cache_file_name(constants::KEY_LCP, config);
osfstream lcp_out_buf(lcp_file, std::ios::binary | std::ios::app | std::ios::out); // open buffer for lcp
size_type bit_size = n*lcp_width;
lcp_out_buf.write((char*) &(bit_size), sizeof(bit_size)); // write size of vector
lcp_out_buf.write((char*) &(lcp_width),sizeof(lcp_width)); // write int_width of vector
size_type wb = 0; // bytes written into lcp int_vector
size_type buffer_size = 1000000; // buffer_size is a multiple of 8!
int_vector<> lcp_buf(buffer_size, 0, lcp_width);
lcp_buf[0] = 0;
sa_buf.reset(buffer_size);
size_type r = 0;// sa_buf.load_next_block();
for (size_type i=1, r_sum=0; r_sum < n;) {
for (; i < r_sum+r; ++i) {
size_type sai = sa_buf[i-r_sum];
lcp_buf[ i-r_sum ] = plcp[sai];
}
if (r > 0) {
size_type cur_wb = (r*lcp_buf.width()+7)/8;
lcp_out_buf.write((const char*)lcp_buf.data(), cur_wb);
wb += cur_wb;
}
r_sum += r; r = sa_buf.load_next_block();
}
if (wb%8) {
lcp_out_buf.write("\0\0\0\0\0\0\0\0", 8-wb%8);
}
lcp_out_buf.close();
register_cache_file(constants::KEY_LCP, config);
}