summaryrefslogtreecommitdiff
path: root/gi/pf
diff options
context:
space:
mode:
authorChris Dyer <cdyer@cs.cmu.edu>2011-12-29 23:02:50 -0500
committerChris Dyer <cdyer@cs.cmu.edu>2011-12-29 23:02:50 -0500
commit031dc91814c1b57269b8a789c93aad0da0a46b6b (patch)
tree212c04385b6bdf65679b0a8c1ba87a168bb74b6e /gi/pf
parent68870390f6c6429af2b9c182ad28e8a78a0f1752 (diff)
remove broken prior, add logging
Diffstat (limited to 'gi/pf')
-rw-r--r--gi/pf/align-lexonly.cc36
1 files changed, 2 insertions, 34 deletions
diff --git a/gi/pf/align-lexonly.cc b/gi/pf/align-lexonly.cc
index 91a3cfcf..7e48b25a 100644
--- a/gi/pf/align-lexonly.cc
+++ b/gi/pf/align-lexonly.cc
@@ -66,41 +66,9 @@ struct AlignedSentencePair {
Array2D<short> posterior;
};
-struct HierarchicalUnigramBase {
- explicit HierarchicalUnigramBase(const unsigned vocab_e_size) : r(5,5), u0(1.0 / vocab_e_size) {}
-
- // return p0 of rule.e_
- prob_t operator()(const TRule& rule) const {
- prob_t p = prob_t::One();
- prob_t q;
- for (unsigned i = 0; i < rule.e_.size(); ++i) {
- q.logeq(r.logprob(rule.e_[i], log(u0)));
- p *= q;
- }
- q.logeq(r.logprob(TD::Convert("</s>"), log(u0)));
- p *= q;
- return p;
- }
-
- void Increment(const TRule& rule) {
- for (unsigned i = 0; i < rule.e_.size(); ++i)
- r.increment(rule.e_[i]);
- r.increment(TD::Convert("</s>"));
- }
-
- void Decrement(const TRule& rule) {
- for (unsigned i = 0; i < rule.e_.size(); ++i)
- r.decrement(rule.e_[i]);
- r.decrement(TD::Convert("</s>"));
- }
-
- CCRP_NoTable<WordID> r;
- prob_t u0;
-};
-
struct HierarchicalWordBase {
explicit HierarchicalWordBase(const unsigned vocab_e_size) :
- base(prob_t::One()), r(15,15), u0(-log(vocab_e_size)) {}
+ base(prob_t::One()), r(25,25,10), u0(-log(vocab_e_size)) {}
void ResampleHyperparameters(MT19937* rng) {
r.resample_hyperparameters(rng);
@@ -137,7 +105,7 @@ struct HierarchicalWordBase {
}
void Summary() const {
- cerr << "NUMBER OF CUSTOMERS: " << r.num_customers() << endl;
+ cerr << "NUMBER OF CUSTOMERS: " << r.num_customers() << " (\\alpha=" << r.concentration() << ')' << endl;
for (CCRP_NoTable<vector<WordID> >::const_iterator it = r.begin(); it != r.end(); ++it)
cerr << " " << it->second << '\t' << TD::GetString(it->first) << endl;
}