std::vector<std::pair<doc_id, double>>
ranker::score(inverted_index& idx, corpus::document& query,
uint64_t num_results /* = 10 */,
const std::function<bool(doc_id d_id)>& filter /* return true */)
{
if (query.counts().empty())
idx.tokenize(query);
score_data sd{idx, idx.avg_doc_length(),
idx.num_docs(), idx.total_corpus_terms(),
query};
// zeros out elements and (if necessary) resizes the vector; this eliminates
// constructing a new vector each query for the same index
results_.assign(sd.num_docs, std::numeric_limits<double>::lowest());
for (auto& tpair : query.counts())
{
term_id t_id{idx.get_term_id(tpair.first)};
auto pdata = idx.search_primary(t_id);
sd.doc_count = pdata->counts().size();
sd.t_id = t_id;
sd.query_term_count = tpair.second;
sd.corpus_term_count = idx.total_num_occurences(sd.t_id);
for (auto& dpair : pdata->counts())
{
sd.d_id = dpair.first;
sd.doc_term_count = dpair.second;
sd.doc_size = idx.doc_size(dpair.first);
sd.doc_unique_terms = idx.unique_terms(dpair.first);
// if this is the first time we've seen this document, compute
// its initial score
if (results_[dpair.first] == std::numeric_limits<double>::lowest())
results_[dpair.first] = initial_score(sd);
results_[dpair.first] += score_one(sd);
}
}
using doc_pair = std::pair<doc_id, double>;
auto doc_pair_comp = [](const doc_pair& a, const doc_pair& b)
{ return a.second > b.second; };
std::priority_queue<doc_pair,
std::vector<doc_pair>,
decltype(doc_pair_comp)> pq{doc_pair_comp};
for (uint64_t id = 0; id < results_.size(); ++id)
{
if (!filter(doc_id{id}))
continue;
pq.emplace(doc_id{id}, results_[id]);
if (pq.size() > num_results)
pq.pop();
}
std::vector<doc_pair> sorted;
while (!pq.empty())
{
sorted.emplace_back(pq.top());
pq.pop();
}
std::reverse(sorted.begin(), sorted.end());
return sorted;
}
std::vector<search_result> ranker::rank(detail::ranker_context& ctx,
uint64_t num_results,
const filter_function_type& filter)
{
score_data sd{ctx.idx, ctx.idx.avg_doc_length(), ctx.idx.num_docs(),
ctx.idx.total_corpus_terms(), ctx.query_length};
auto comp = [](const search_result& a, const search_result& b)
{
// comparison is reversed since we want a min-heap
return a.score > b.score;
};
util::fixed_heap<search_result, decltype(comp)> results{num_results, comp};
doc_id next_doc{ctx.idx.num_docs()};
while (ctx.cur_doc < ctx.idx.num_docs())
{
sd.d_id = ctx.cur_doc;
sd.doc_size = ctx.idx.doc_size(ctx.cur_doc);
sd.doc_unique_terms = ctx.idx.unique_terms(ctx.cur_doc);
auto score = initial_score(sd);
for (auto& pc : ctx.postings)
{
if (pc.begin == pc.end)
continue;
if (pc.begin->first == ctx.cur_doc)
{
// set up this term
sd.t_id = pc.t_id;
sd.query_term_weight = pc.query_term_weight;
sd.doc_count = pc.doc_count;
sd.corpus_term_count = pc.corpus_term_count;
sd.doc_term_count = pc.begin->second;
score += score_one(sd);
// advance over this position in the current postings context
// until the next valid document
do
{
++pc.begin;
} while (pc.begin != pc.end && !filter(pc.begin->first));
}
if (pc.begin != pc.end)
{
// check if the document in the next position is the
// smallest accepted doc_id
if (pc.begin->first < next_doc)
next_doc = pc.begin->first;
}
}
results.emplace(ctx.cur_doc, score);
ctx.cur_doc = next_doc;
next_doc = doc_id{ctx.idx.num_docs()};
}
return results.extract_top();
}
