std::string get_content(const corpus::document& doc)
{
if (!doc.contains_content())
throw analyzer_exception{
"document content was not populated for analysis"};
return utf::to_utf8(doc.content(), doc.encoding());
}
void depth_featurizer::tree_tokenize(corpus::document& doc,
const parser::parse_tree& tree) const
{
height_visitor vtor;
auto rep = "depth-" + std::to_string(tree.visit(vtor));
doc.increment(rep, 1);
}
void depth_analyzer::tree_tokenize(corpus::document& doc,
const parse_tree& tree)
{
size_t h = parse_tree::height(tree);
std::string representation = std::to_string(h);
doc.increment(representation, 1);
}
void check_analyzer_expected(Analyzer& ana, corpus::document doc,
uint64_t num_unique, uint64_t length)
{
ana.tokenize(doc);
ASSERT_EQUAL(doc.counts().size(), num_unique);
ASSERT_EQUAL(doc.length(), length);
ASSERT_EQUAL(doc.id(), 47ul);
if (doc.contains_content())
{
ASSERT_EQUAL(doc.path(), "/home/person/filename.txt");
ASSERT_EQUAL(doc.name(), "filename.txt");
}
else
{
ASSERT_EQUAL(doc.path(), "../data/sample-document.txt");
ASSERT_EQUAL(doc.name(), "sample-document.txt");
}
}
void ngram_pos_analyzer::tokenize(corpus::document& doc)
{
// first, get tokens
stream_->set_content(get_content(doc));
std::vector<sequence::sequence> sentences;
sequence::sequence seq;
// put tokens into sequences, excluding sentence markers
while (*stream_)
{
auto next = stream_->next();
if (next.empty() || next == " " || next == "<s>")
continue;
if (next == "</s>")
sentences.emplace_back(std::move(seq));
else
seq.add_observation(
{sequence::symbol_t{next}, sequence::tag_t{"[unknown]"}});
}
auto tagger = crf_->make_tagger();
for (auto& seq : sentences)
{
// generate CRF features
seq_analyzer_.analyze(seq);
// POS-tag sentence
tagger.tag(seq);
// create ngrams
for (size_t i = n_value() - 1; i < seq.size(); ++i)
{
std::string combined = seq_analyzer_.tag(seq[i].label());
for (size_t j = 1; j < n_value(); ++j)
{
std::string next = seq_analyzer_.tag(seq[i - j].label());
combined = next + "_" + combined;
}
doc.increment(combined, 1);
}
}
}
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;
}