From 4c2360119def2fb624d2691b355b1908c511f004 Mon Sep 17 00:00:00 2001
From: Chris Dyer <cdyer@cs.cmu.edu>
Date: Tue, 24 Jan 2012 22:26:44 -0500
Subject: more models
---
gi/pf/align-lexonly.cc | 14 ++++++++++----
gi/pf/base_measures.cc | 2 +-
gi/pf/base_measures.h | 27 ++++++++++++++++++++++++++-
3 files changed, 37 insertions(+), 6 deletions(-)
(limited to 'gi/pf')
diff --git a/gi/pf/align-lexonly.cc b/gi/pf/align-lexonly.cc
index e9f1e7b6..76e2e009 100644
--- a/gi/pf/align-lexonly.cc
+++ b/gi/pf/align-lexonly.cc
@@ -122,10 +122,11 @@ struct BasicLexicalAlignment {
vector<AlignedSentencePair>* corp) :
letters(lets),
corpus(*corp),
+ up0("fr-en.10k.translit-base.txt.gz"),
//up0(words_e),
//up0("en.chars.1gram", letters_e),
//up0("en.words.1gram"),
- up0(letters_e),
+ //up0(letters_e),
//up0("en.chars.2gram"),
tmodel(up0) {
}
@@ -180,14 +181,18 @@ struct BasicLexicalAlignment {
//PhraseConditionalUninformativeUnigramBase up0;
//UnigramWordBase up0;
//HierarchicalUnigramBase up0;
- HierarchicalWordBase up0;
+ TableLookupBase up0;
+ //HierarchicalWordBase up0;
+ //PoissonUniformUninformativeBase up0;
//CompletelyUniformBase up0;
//FixedNgramBase up0;
//ConditionalTranslationModel<PhraseConditionalUninformativeBase> tmodel;
//ConditionalTranslationModel<PhraseConditionalUninformativeUnigramBase> tmodel;
//ConditionalTranslationModel<UnigramWordBase> tmodel;
//ConditionalTranslationModel<HierarchicalUnigramBase> tmodel;
- ConditionalTranslationModel<HierarchicalWordBase> tmodel;
+ //ConditionalTranslationModel<HierarchicalWordBase> tmodel;
+ //ConditionalTranslationModel<PoissonUniformUninformativeBase> tmodel;
+ ConditionalTranslationModel<TableLookupBase> tmodel;
//ConditionalTranslationModel<FixedNgramBase> tmodel;
//ConditionalTranslationModel<CompletelyUniformBase> tmodel;
};
@@ -222,6 +227,7 @@ void BasicLexicalAlignment::ResampleCorpus() {
void ExtractLetters(const set<WordID>& v, vector<vector<WordID> >* l, set<WordID>* letset = NULL) {
for (set<WordID>::const_iterator it = v.begin(); it != v.end(); ++it) {
+ if (*it >= l->size()) { l->resize(*it + 1); }
vector<WordID>& letters = (*l)[*it];
if (letters.size()) continue; // if e and f have the same word
@@ -308,7 +314,7 @@ int main(int argc, char** argv) {
x.InitializeRandom();
const unsigned samples = conf["samples"].as<unsigned>();
for (int i = 0; i < samples; ++i) {
- for (int j = 4995; j < 4997; ++j) Debug(corpus[j]);
+ for (int j = 395; j < 397; ++j) Debug(corpus[j]);
cerr << i << "\t" << x.tmodel.r.size() << "\t";
if (i % 10 == 0) x.ResampleHyperparemeters();
x.ResampleCorpus();
diff --git a/gi/pf/base_measures.cc b/gi/pf/base_measures.cc
index 7894d3e7..4b1863fa 100644
--- a/gi/pf/base_measures.cc
+++ b/gi/pf/base_measures.cc
@@ -37,7 +37,7 @@ TableLookupBase::TableLookupBase(const string& fname) {
} else if (cc == 1) {
x.e_.push_back(cur);
} else if (cc == 2) {
- table[x] = atof(TD::Convert(cur));
+ table[x].logeq(atof(TD::Convert(cur)));
++cc;
} else {
if (flag) cerr << endl;
diff --git a/gi/pf/base_measures.h b/gi/pf/base_measures.h
index 7214aa22..b0495bfd 100644
--- a/gi/pf/base_measures.h
+++ b/gi/pf/base_measures.h
@@ -51,6 +51,22 @@ struct Model1 {
std::vector<std::map<WordID, prob_t> > ttable;
};
+struct PoissonUniformUninformativeBase {
+ explicit PoissonUniformUninformativeBase(const unsigned ves) : kUNIFORM(1.0 / ves) {}
+ prob_t operator()(const TRule& r) const {
+ prob_t p; p.logeq(log_poisson(r.e_.size(), 1.0));
+ prob_t q = kUNIFORM; q.poweq(r.e_.size());
+ p *= q;
+ return p;
+ }
+ void Summary() const {}
+ void ResampleHyperparameters(MT19937*) {}
+ void Increment(const TRule&) {}
+ void Decrement(const TRule&) {}
+ prob_t Likelihood() const { return prob_t::One(); }
+ const prob_t kUNIFORM;
+};
+
struct CompletelyUniformBase {
explicit CompletelyUniformBase(const unsigned ves) : kUNIFORM(1.0 / ves) {}
prob_t operator()(const TRule&) const {
@@ -83,10 +99,19 @@ struct TableLookupBase {
prob_t operator()(const TRule& rule) const {
const std::tr1::unordered_map<TRule,prob_t>::const_iterator it = table.find(rule);
- assert(it != table.end());
+ if (it == table.end()) {
+ std::cerr << rule << " not found\n";
+ abort();
+ }
return it->second;
}
+ void ResampleHyperparameters(MT19937*) {}
+ void Increment(const TRule&) {}
+ void Decrement(const TRule&) {}
+ prob_t Likelihood() const { return prob_t::One(); }
+ void Summary() const {}
+
std::tr1::unordered_map<TRule,prob_t,RuleHasher> table;
};
--
cgit v1.2.3