summaryrefslogtreecommitdiff
path: root/utils
diff options
context:
space:
mode:
authorPatrick Simianer <p@simianer.de>2011-10-19 14:02:34 +0200
committerPatrick Simianer <p@simianer.de>2011-10-19 14:02:34 +0200
commit9beaeb42b71fa504bfa41a402cb17eb6ac4001af (patch)
tree0add4afabc526391753e4e6b9443a7bf21e3e2c3 /utils
parentce3b4db94d40c111ede321ac6de2bb061a81c4af (diff)
parent09297047e446f49804d3f48bf320cdbd38d6396a (diff)
merge upstream/master
Diffstat (limited to 'utils')
-rw-r--r--utils/Makefile.am14
-rw-r--r--utils/ccrp_nt.h169
-rw-r--r--utils/ccrp_onetable.h241
-rw-r--r--utils/fdict.cc4
-rw-r--r--utils/fdict.h41
-rwxr-xr-xutils/feature_vector.h4
-rw-r--r--utils/filelib.cc31
-rw-r--r--utils/filelib.h6
-rw-r--r--utils/logval.h10
-rw-r--r--utils/logval_test.cc14
-rw-r--r--utils/perfect_hash.cc37
-rw-r--r--utils/perfect_hash.h24
-rw-r--r--utils/phmt.cc40
-rw-r--r--utils/reconstruct_weights.cc68
-rw-r--r--utils/sampler.h2
-rw-r--r--utils/sparse_vector.h38
-rw-r--r--utils/stringlib.cc370
-rw-r--r--utils/stringlib.h3
-rw-r--r--utils/tdict.cc4
-rw-r--r--utils/ts.cc6
-rw-r--r--utils/weights.cc196
-rw-r--r--utils/weights.h30
-rw-r--r--utils/weights_test.cc7
23 files changed, 1204 insertions, 155 deletions
diff --git a/utils/Makefile.am b/utils/Makefile.am
index 94f9be30..df667655 100644
--- a/utils/Makefile.am
+++ b/utils/Makefile.am
@@ -1,5 +1,8 @@
-noinst_PROGRAMS = ts
-TESTS = ts
+
+bin_PROGRAMS = reconstruct_weights
+
+noinst_PROGRAMS = ts phmt
+TESTS = ts phmt
if HAVE_GTEST
noinst_PROGRAMS += \
@@ -11,6 +14,8 @@ noinst_PROGRAMS += \
TESTS += small_vector_test logval_test weights_test dict_test
endif
+reconstruct_weights_SOURCES = reconstruct_weights.cc
+
noinst_LIBRARIES = libutils.a
libutils_a_SOURCES = \
@@ -27,6 +32,11 @@ libutils_a_SOURCES = \
verbose.cc \
weights.cc
+if HAVE_CMPH
+ libutils_a_SOURCES += perfect_hash.cc
+endif
+
+phmt_SOURCES = phmt.cc
ts_SOURCES = ts.cc
dict_test_SOURCES = dict_test.cc
dict_test_LDADD = $(GTEST_LDFLAGS) $(GTEST_LIBS)
diff --git a/utils/ccrp_nt.h b/utils/ccrp_nt.h
new file mode 100644
index 00000000..63b6f4c2
--- /dev/null
+++ b/utils/ccrp_nt.h
@@ -0,0 +1,169 @@
+#ifndef _CCRP_NT_H_
+#define _CCRP_NT_H_
+
+#include <numeric>
+#include <cassert>
+#include <cmath>
+#include <list>
+#include <iostream>
+#include <vector>
+#include <tr1/unordered_map>
+#include <boost/functional/hash.hpp>
+#include "sampler.h"
+#include "slice_sampler.h"
+
+// Chinese restaurant process (1 parameter)
+template <typename Dish, typename DishHash = boost::hash<Dish> >
+class CCRP_NoTable {
+ public:
+ explicit CCRP_NoTable(double conc) :
+ num_customers_(),
+ concentration_(conc),
+ concentration_prior_shape_(std::numeric_limits<double>::quiet_NaN()),
+ concentration_prior_rate_(std::numeric_limits<double>::quiet_NaN()) {}
+
+ CCRP_NoTable(double c_shape, double c_rate, double c = 10.0) :
+ num_customers_(),
+ concentration_(c),
+ concentration_prior_shape_(c_shape),
+ concentration_prior_rate_(c_rate) {}
+
+ double concentration() const { return concentration_; }
+
+ bool has_concentration_prior() const {
+ return !std::isnan(concentration_prior_shape_);
+ }
+
+ void clear() {
+ num_customers_ = 0;
+ custs_.clear();
+ }
+
+ unsigned num_customers() const {
+ return num_customers_;
+ }
+
+ unsigned num_customers(const Dish& dish) const {
+ const typename std::tr1::unordered_map<Dish, unsigned, DishHash>::const_iterator it = custs_.find(dish);
+ if (it == custs_.end()) return 0;
+ return it->second;
+ }
+
+ int increment(const Dish& dish) {
+ int table_diff = 0;
+ if (++custs_[dish] == 1)
+ table_diff = 1;
+ ++num_customers_;
+ return table_diff;
+ }
+
+ int decrement(const Dish& dish) {
+ int table_diff = 0;
+ int nc = --custs_[dish];
+ if (nc == 0) {
+ custs_.erase(dish);
+ table_diff = -1;
+ } else if (nc < 0) {
+ std::cerr << "Dish counts dropped below zero for: " << dish << std::endl;
+ abort();
+ }
+ --num_customers_;
+ return table_diff;
+ }
+
+ double prob(const Dish& dish, const double& p0) const {
+ const unsigned at_table = num_customers(dish);
+ return (at_table + p0 * concentration_) / (num_customers_ + concentration_);
+ }
+
+ double logprob(const Dish& dish, const double& logp0) const {
+ const unsigned at_table = num_customers(dish);
+ return log(at_table + exp(logp0 + log(concentration_))) - log(num_customers_ + concentration_);
+ }
+
+ double log_crp_prob() const {
+ return log_crp_prob(concentration_);
+ }
+
+ static double log_gamma_density(const double& x, const double& shape, const double& rate) {
+ assert(x >= 0.0);
+ assert(shape > 0.0);
+ assert(rate > 0.0);
+ const double lp = (shape-1)*log(x) - shape*log(rate) - x/rate - lgamma(shape);
+ return lp;
+ }
+
+ // taken from http://en.wikipedia.org/wiki/Chinese_restaurant_process
+ // does not include P_0's
+ double log_crp_prob(const double& concentration) const {
+ double lp = 0.0;
+ if (has_concentration_prior())
+ lp += log_gamma_density(concentration, concentration_prior_shape_, concentration_prior_rate_);
+ assert(lp <= 0.0);
+ if (num_customers_) {
+ lp += lgamma(concentration) - lgamma(concentration + num_customers_) +
+ custs_.size() * log(concentration);
+ assert(std::isfinite(lp));
+ for (typename std::tr1::unordered_map<Dish, unsigned, DishHash>::const_iterator it = custs_.begin();
+ it != custs_.end(); ++it) {
+ lp += lgamma(it->second);
+ }
+ }
+ assert(std::isfinite(lp));
+ return lp;
+ }
+
+ void resample_hyperparameters(MT19937* rng, const unsigned nloop = 5, const unsigned niterations = 10) {
+ assert(has_concentration_prior());
+ ConcentrationResampler cr(*this);
+ for (int iter = 0; iter < nloop; ++iter) {
+ concentration_ = slice_sampler1d(cr, concentration_, *rng, 0.0,
+ std::numeric_limits<double>::infinity(), 0.0, niterations, 100*niterations);
+ }
+ }
+
+ struct ConcentrationResampler {
+ ConcentrationResampler(const CCRP_NoTable& crp) : crp_(crp) {}
+ const CCRP_NoTable& crp_;
+ double operator()(const double& proposed_concentration) const {
+ return crp_.log_crp_prob(proposed_concentration);
+ }
+ };
+
+ void Print(std::ostream* out) const {
+ (*out) << "DP(alpha=" << concentration_ << ") customers=" << num_customers_ << std::endl;
+ int cc = 0;
+ for (typename std::tr1::unordered_map<Dish, unsigned, DishHash>::const_iterator it = custs_.begin();
+ it != custs_.end(); ++it) {
+ (*out) << " " << it->first << "(" << it->second << " eating)";
+ ++cc;
+ if (cc > 10) { (*out) << " ..."; break; }
+ }
+ (*out) << std::endl;
+ }
+
+ unsigned num_customers_;
+ std::tr1::unordered_map<Dish, unsigned, DishHash> custs_;
+
+ typedef typename std::tr1::unordered_map<Dish, unsigned, DishHash>::const_iterator const_iterator;
+ const_iterator begin() const {
+ return custs_.begin();
+ }
+ const_iterator end() const {
+ return custs_.end();
+ }
+
+ double concentration_;
+
+ // optional gamma prior on concentration_ (NaN if no prior)
+ double concentration_prior_shape_;
+ double concentration_prior_rate_;
+};
+
+template <typename T,typename H>
+std::ostream& operator<<(std::ostream& o, const CCRP_NoTable<T,H>& c) {
+ c.Print(&o);
+ return o;
+}
+
+#endif
diff --git a/utils/ccrp_onetable.h b/utils/ccrp_onetable.h
new file mode 100644
index 00000000..a868af9a
--- /dev/null
+++ b/utils/ccrp_onetable.h
@@ -0,0 +1,241 @@
+#ifndef _CCRP_ONETABLE_H_
+#define _CCRP_ONETABLE_H_
+
+#include <numeric>
+#include <cassert>
+#include <cmath>
+#include <list>
+#include <iostream>
+#include <tr1/unordered_map>
+#include <boost/functional/hash.hpp>
+#include "sampler.h"
+#include "slice_sampler.h"
+
+// Chinese restaurant process (Pitman-Yor parameters) with one table approximation
+
+template <typename Dish, typename DishHash = boost::hash<Dish> >
+class CCRP_OneTable {
+ typedef std::tr1::unordered_map<Dish, unsigned, DishHash> DishMapType;
+ public:
+ CCRP_OneTable(double disc, double conc) :
+ num_tables_(),
+ num_customers_(),
+ discount_(disc),
+ concentration_(conc),
+ discount_prior_alpha_(std::numeric_limits<double>::quiet_NaN()),
+ discount_prior_beta_(std::numeric_limits<double>::quiet_NaN()),
+ concentration_prior_shape_(std::numeric_limits<double>::quiet_NaN()),
+ concentration_prior_rate_(std::numeric_limits<double>::quiet_NaN()) {}
+
+ CCRP_OneTable(double d_alpha, double d_beta, double c_shape, double c_rate, double d = 0.9, double c = 1.0) :
+ num_tables_(),
+ num_customers_(),
+ discount_(d),
+ concentration_(c),
+ discount_prior_alpha_(d_alpha),
+ discount_prior_beta_(d_beta),
+ concentration_prior_shape_(c_shape),
+ concentration_prior_rate_(c_rate) {}
+
+ double discount() const { return discount_; }
+ double concentration() const { return concentration_; }
+ void set_concentration(double c) { concentration_ = c; }
+ void set_discount(double d) { discount_ = d; }
+
+ bool has_discount_prior() const {
+ return !std::isnan(discount_prior_alpha_);
+ }
+
+ bool has_concentration_prior() const {
+ return !std::isnan(concentration_prior_shape_);
+ }
+
+ void clear() {
+ num_tables_ = 0;
+ num_customers_ = 0;
+ dish_counts_.clear();
+ }
+
+ unsigned num_tables() const {
+ return num_tables_;
+ }
+
+ unsigned num_tables(const Dish& dish) const {
+ const typename DishMapType::const_iterator it = dish_counts_.find(dish);
+ if (it == dish_counts_.end()) return 0;
+ return 1;
+ }
+
+ unsigned num_customers() const {
+ return num_customers_;
+ }
+
+ unsigned num_customers(const Dish& dish) const {
+ const typename DishMapType::const_iterator it = dish_counts_.find(dish);
+ if (it == dish_counts_.end()) return 0;
+ return it->second;
+ }
+
+ // returns +1 or 0 indicating whether a new table was opened
+ int increment(const Dish& dish) {
+ unsigned& dc = dish_counts_[dish];
+ ++dc;
+ ++num_customers_;
+ if (dc == 1) {
+ ++num_tables_;
+ return 1;
+ } else {
+ return 0;
+ }
+ }
+
+ // returns -1 or 0, indicating whether a table was closed
+ int decrement(const Dish& dish) {
+ unsigned& dc = dish_counts_[dish];
+ assert(dc > 0);
+ if (dc == 1) {
+ dish_counts_.erase(dish);
+ --num_tables_;
+ --num_customers_;
+ return -1;
+ } else {
+ assert(dc > 1);
+ --dc;
+ --num_customers_;
+ return 0;
+ }
+ }
+
+ double prob(const Dish& dish, const double& p0) const {
+ const typename DishMapType::const_iterator it = dish_counts_.find(dish);
+ const double r = num_tables_ * discount_ + concentration_;
+ if (it == dish_counts_.end()) {
+ return r * p0 / (num_customers_ + concentration_);
+ } else {
+ return (it->second - discount_ + r * p0) /
+ (num_customers_ + concentration_);
+ }
+ }
+
+ double log_crp_prob() const {
+ return log_crp_prob(discount_, concentration_);
+ }
+
+ static double log_beta_density(const double& x, const double& alpha, const double& beta) {
+ assert(x > 0.0);
+ assert(x < 1.0);
+ assert(alpha > 0.0);
+ assert(beta > 0.0);
+ const double lp = (alpha-1)*log(x)+(beta-1)*log(1-x)+lgamma(alpha+beta)-lgamma(alpha)-lgamma(beta);
+ return lp;
+ }
+
+ static double log_gamma_density(const double& x, const double& shape, const double& rate) {
+ assert(x >= 0.0);
+ assert(shape > 0.0);
+ assert(rate > 0.0);
+ const double lp = (shape-1)*log(x) - shape*log(rate) - x/rate - lgamma(shape);
+ return lp;
+ }
+
+ // taken from http://en.wikipedia.org/wiki/Chinese_restaurant_process
+ // does not include P_0's
+ double log_crp_prob(const double& discount, const double& concentration) const {
+ double lp = 0.0;
+ if (has_discount_prior())
+ lp = log_beta_density(discount, discount_prior_alpha_, discount_prior_beta_);
+ if (has_concentration_prior())
+ lp += log_gamma_density(concentration, concentration_prior_shape_, concentration_prior_rate_);
+ assert(lp <= 0.0);
+ if (num_customers_) {
+ if (discount > 0.0) {
+ const double r = lgamma(1.0 - discount);
+ lp += lgamma(concentration) - lgamma(concentration + num_customers_)
+ + num_tables_ * log(discount) + lgamma(concentration / discount + num_tables_)
+ - lgamma(concentration / discount);
+ assert(std::isfinite(lp));
+ for (typename DishMapType::const_iterator it = dish_counts_.begin();
+ it != dish_counts_.end(); ++it) {
+ const unsigned& cur = it->second;
+ lp += lgamma(cur - discount) - r;
+ }
+ } else {
+ assert(!"not implemented yet");
+ }
+ }
+ assert(std::isfinite(lp));
+ return lp;
+ }
+
+ void resample_hyperparameters(MT19937* rng, const unsigned nloop = 5, const unsigned niterations = 10) {
+ assert(has_discount_prior() || has_concentration_prior());
+ DiscountResampler dr(*this);
+ ConcentrationResampler cr(*this);
+ for (int iter = 0; iter < nloop; ++iter) {
+ if (has_concentration_prior()) {
+ concentration_ = slice_sampler1d(cr, concentration_, *rng, 0.0,
+ std::numeric_limits<double>::infinity(), 0.0, niterations, 100*niterations);
+ }
+ if (has_discount_prior()) {
+ discount_ = slice_sampler1d(dr, discount_, *rng, std::numeric_limits<double>::min(),
+ 1.0, 0.0, niterations, 100*niterations);
+ }
+ }
+ concentration_ = slice_sampler1d(cr, concentration_, *rng, 0.0,
+ std::numeric_limits<double>::infinity(), 0.0, niterations, 100*niterations);
+ }
+
+ struct DiscountResampler {
+ DiscountResampler(const CCRP_OneTable& crp) : crp_(crp) {}
+ const CCRP_OneTable& crp_;
+ double operator()(const double& proposed_discount) const {
+ return crp_.log_crp_prob(proposed_discount, crp_.concentration_);
+ }
+ };
+
+ struct ConcentrationResampler {
+ ConcentrationResampler(const CCRP_OneTable& crp) : crp_(crp) {}
+ const CCRP_OneTable& crp_;
+ double operator()(const double& proposed_concentration) const {
+ return crp_.log_crp_prob(crp_.discount_, proposed_concentration);
+ }
+ };
+
+ void Print(std::ostream* out) const {
+ (*out) << "PYP(d=" << discount_ << ",c=" << concentration_ << ") customers=" << num_customers_ << std::endl;
+ for (typename DishMapType::const_iterator it = dish_counts_.begin(); it != dish_counts_.end(); ++it) {
+ (*out) << " " << it->first << " = " << it->second << std::endl;
+ }
+ }
+
+ typedef typename DishMapType::const_iterator const_iterator;
+ const_iterator begin() const {
+ return dish_counts_.begin();
+ }
+ const_iterator end() const {
+ return dish_counts_.end();
+ }
+
+ unsigned num_tables_;
+ unsigned num_customers_;
+ DishMapType dish_counts_;
+
+ double discount_;
+ double concentration_;
+
+ // optional beta prior on discount_ (NaN if no prior)
+ double discount_prior_alpha_;
+ double discount_prior_beta_;
+
+ // optional gamma prior on concentration_ (NaN if no prior)
+ double concentration_prior_shape_;
+ double concentration_prior_rate_;
+};
+
+template <typename T,typename H>
+std::ostream& operator<<(std::ostream& o, const CCRP_OneTable<T,H>& c) {
+ c.Print(&o);
+ return o;
+}
+
+#endif
diff --git a/utils/fdict.cc b/utils/fdict.cc
index baa0b552..676c951c 100644
--- a/utils/fdict.cc
+++ b/utils/fdict.cc
@@ -9,6 +9,10 @@ using namespace std;
Dict FD::dict_;
bool FD::frozen_ = false;
+#ifdef HAVE_CMPH
+PerfectHashFunction* FD::hash_ = NULL;
+#endif
+
std::string FD::Convert(std::vector<WordID> const& v) {
return Convert(&*v.begin(),&*v.end());
}
diff --git a/utils/fdict.h b/utils/fdict.h
index 70315a38..9c8d7cde 100644
--- a/utils/fdict.h
+++ b/utils/fdict.h
@@ -1,27 +1,59 @@
#ifndef _FDICT_H_
#define _FDICT_H_
+#include "config.h"
+
+#include <iostream>
#include <string>
#include <vector>
#include "dict.h"
+#ifdef HAVE_CMPH
+#include "perfect_hash.h"
+#include "string_to.h"
+#endif
+
struct FD {
// once the FD is frozen, new features not already in the
// dictionary will return 0
static void Freeze() {
frozen_ = true;
}
- static void UnFreeze() {
- frozen_ = false;
+ static bool UsingPerfectHashFunction() {
+#ifdef HAVE_CMPH
+ return hash_;
+#else
+ return false;
+#endif
}
-
+ static void EnableHash(const std::string& cmph_file) {
+#ifdef HAVE_CMPH
+ assert(dict_.max() == 0); // dictionary must not have
+ // been added to
+ hash_ = new PerfectHashFunction(cmph_file);
+#endif
+ }
+>>>>>>> upstream/master
static inline int NumFeats() {
+#ifdef HAVE_CMPH
+ if (hash_) return hash_->number_of_keys();
+#endif
return dict_.max() + 1;
}
static inline WordID Convert(const std::string& s) {
+#ifdef HAVE_CMPH
+ if (hash_) return (*hash_)(s);
+#endif
return dict_.Convert(s, frozen_);
}
static inline const std::string& Convert(const WordID& w) {
+#ifdef HAVE_CMPH
+ if (hash_) {
+ static std::string tls;
+ tls = to_string(w);
+ return tls;
+ }
+#endif
return dict_.Convert(w);
}
static std::string Convert(WordID const *i,WordID const* e);
@@ -33,6 +65,9 @@ struct FD {
static Dict dict_;
private:
static bool frozen_;
+#ifdef HAVE_CMPH
+ static PerfectHashFunction* hash_;
+#endif
};
#endif
diff --git a/utils/feature_vector.h b/utils/feature_vector.h
index 733aa99e..a7b61a66 100755
--- a/utils/feature_vector.h
+++ b/utils/feature_vector.h
@@ -3,9 +3,9 @@
#include <vector>
#include "sparse_vector.h"
-#include "fdict.h"
+#include "weights.h"
-typedef double Featval;
+typedef weight_t Featval;
typedef SparseVector<Featval> FeatureVector;
typedef SparseVector<Featval> WeightVector;
typedef std::vector<Featval> DenseWeightVector;
diff --git a/utils/filelib.cc b/utils/filelib.cc
index 79ad2847..d206fc19 100644
--- a/utils/filelib.cc
+++ b/utils/filelib.cc
@@ -2,6 +2,12 @@
#include <unistd.h>
#include <sys/stat.h>
+#include <sys/types.h>
+#include <sys/socket.h>
+#include <cstdlib>
+#include <cstdio>
+#include <sys/stat.h>
+#include <sys/types.h>
using namespace std;
@@ -20,3 +26,28 @@ bool DirectoryExists(const string& dir) {
return false;
}
+void MkDirP(const string& dir) {
+ if (DirectoryExists(dir)) return;
+ if (mkdir(dir.c_str(), 0777)) {
+ perror(dir.c_str());
+ abort();
+ }
+ if (chmod(dir.c_str(), 07777)) {
+ perror(dir.c_str());
+ abort();
+ }
+}
+
+#if 0
+void CopyFile(const string& inf, const string& outf) {
+ WriteFile w(outf);
+ CopyFile(inf,*w);
+}
+#else
+void CopyFile(const string& inf, const string& outf) {
+ ofstream of(outf.c_str(), fstream::trunc|fstream::binary);
+ ifstream in(inf.c_str(), fstream::binary);
+ of << in.rdbuf();
+}
+#endif
+
diff --git a/utils/filelib.h b/utils/filelib.h
index dda98671..bb6e7415 100644
--- a/utils/filelib.h
+++ b/utils/filelib.h
@@ -12,6 +12,7 @@
bool FileExists(const std::string& file_name);
bool DirectoryExists(const std::string& dir_name);
+void MkDirP(const std::string& dir_name);
// reads from standard in if filename is -
// uncompresses if file ends with .gz
@@ -112,9 +113,6 @@ inline void CopyFile(std::string const& inf,std::ostream &out) {
CopyFile(*r,out);
}
-inline void CopyFile(std::string const& inf,std::string const& outf) {
- WriteFile w(outf);
- CopyFile(inf,*w);
-}
+void CopyFile(std::string const& inf,std::string const& outf);
#endif
diff --git a/utils/logval.h b/utils/logval.h
index 6fdc2c42..8a59d0b1 100644
--- a/utils/logval.h
+++ b/utils/logval.h
@@ -25,12 +25,13 @@ class LogVal {
typedef LogVal<T> Self;
LogVal() : s_(), v_(LOGVAL_LOG0) {}
- explicit LogVal(double x) : s_(std::signbit(x)), v_(s_ ? std::log(-x) : std::log(x)) {}
+ LogVal(double x) : s_(std::signbit(x)), v_(s_ ? std::log(-x) : std::log(x)) {}
+ const Self& operator=(double x) { s_ = std::signbit(x); v_ = s_ ? std::log(-x) : std::log(x); return *this; }
LogVal(init_minus_1) : s_(true),v_(0) { }
LogVal(init_1) : s_(),v_(0) { }
LogVal(init_0) : s_(),v_(LOGVAL_LOG0) { }
- LogVal(int x) : s_(x<0), v_(s_ ? std::log(-x) : std::log(x)) {}
- LogVal(unsigned x) : s_(0), v_(std::log(x)) { }
+ explicit LogVal(int x) : s_(x<0), v_(s_ ? std::log(-x) : std::log(x)) {}
+ explicit LogVal(unsigned x) : s_(0), v_(std::log(x)) { }
LogVal(double lnx,bool sign) : s_(sign),v_(lnx) {}
LogVal(double lnx,init_lnx) : s_(),v_(lnx) {}
static Self exp(T lnx) { return Self(lnx,false); }
@@ -141,9 +142,6 @@ class LogVal {
return pow(1/root);
}
- operator T() const {
- if (s_) return -std::exp(v_); else return std::exp(v_);
- }
T as_float() const {
if (s_) return -std::exp(v_); else return std::exp(v_);
}
diff --git a/utils/logval_test.cc b/utils/logval_test.cc
index 4aa452f2..6133f5ce 100644
--- a/utils/logval_test.cc
+++ b/utils/logval_test.cc
@@ -30,13 +30,13 @@ TEST_F(LogValTest,Negate) {
LogVal<double> x(-2.4);
LogVal<double> y(2.4);
y.negate();
- EXPECT_FLOAT_EQ(x,y);
+ EXPECT_FLOAT_EQ(x.as_float(),y.as_float());
}
TEST_F(LogValTest,Inverse) {
LogVal<double> x(1/2.4);
LogVal<double> y(2.4);
- EXPECT_FLOAT_EQ(x,y.inverse());
+ EXPECT_FLOAT_EQ(x.as_float(),y.inverse().as_float());
}
TEST_F(LogValTest,Minus) {
@@ -45,9 +45,9 @@ TEST_F(LogValTest,Minus) {
LogVal<double> z1 = x - y;
LogVal<double> z2 = x;
z2 -= y;
- EXPECT_FLOAT_EQ(z1, z2);
- EXPECT_FLOAT_EQ(z1, 10.0);
- EXPECT_FLOAT_EQ(y - x, -10.0);
+ EXPECT_FLOAT_EQ(z1.as_float(), z2.as_float());
+ EXPECT_FLOAT_EQ(z1.as_float(), 10.0);
+ EXPECT_FLOAT_EQ((y - x).as_float(), -10.0);
}
TEST_F(LogValTest,TestOps) {
@@ -62,8 +62,8 @@ TEST_F(LogValTest,TestOps) {
LogVal<double> bb(-0.3);
cerr << (aa + bb) << endl;
cerr << (bb + aa) << endl;
- EXPECT_FLOAT_EQ((aa + bb), (bb + aa));
- EXPECT_FLOAT_EQ((aa + bb), -0.1);
+ EXPECT_FLOAT_EQ((aa + bb).as_float(), (bb + aa).as_float());
+ EXPECT_FLOAT_EQ((aa + bb).as_float(), -0.1);
}
TEST_F(LogValTest,TestSizes) {
diff --git a/utils/perfect_hash.cc b/utils/perfect_hash.cc
new file mode 100644
index 00000000..706e2741
--- /dev/null
+++ b/utils/perfect_hash.cc
@@ -0,0 +1,37 @@
+#include "config.h"
+
+#ifdef HAVE_CMPH
+
+#include "perfect_hash.h"
+
+#include <cstdio>
+#include <iostream>
+
+using namespace std;
+
+PerfectHashFunction::~PerfectHashFunction() {
+ cmph_destroy(mphf_);
+}
+
+PerfectHashFunction::PerfectHashFunction(const string& fname) {
+ FILE* f = fopen(fname.c_str(), "r");
+ if (!f) {
+ cerr << "Failed to open file " << fname << " for reading: cannot load hash function.\n";
+ abort();
+ }
+ mphf_ = cmph_load(f);
+ if (!mphf_) {
+ cerr << "cmph_load failed on " << fname << "!\n";
+ abort();
+ }
+}
+
+size_t PerfectHashFunction::operator()(const string& key) const {
+ return cmph_search(mphf_, &key[0], key.size());
+}
+
+size_t PerfectHashFunction::number_of_keys() const {
+ return cmph_size(mphf_);
+}
+
+#endif
diff --git a/utils/perfect_hash.h b/utils/perfect_hash.h
new file mode 100644
index 00000000..8ac11f18
--- /dev/null
+++ b/utils/perfect_hash.h
@@ -0,0 +1,24 @@
+#ifndef _PERFECT_HASH_MAP_H_
+#define _PERFECT_HASH_MAP_H_
+
+#include "config.h"
+
+#ifndef HAVE_CMPH
+#error libcmph is required to use PerfectHashFunction
+#endif
+
+#include <vector>
+#include <boost/utility.hpp>
+#include "cmph.h"
+
+class PerfectHashFunction : boost::noncopyable {
+ public:
+ explicit PerfectHashFunction(const std::string& fname);
+ ~PerfectHashFunction();
+ size_t operator()(const std::string& key) const;
+ size_t number_of_keys() const;
+ private:
+ cmph_t *mphf_;
+};
+
+#endif
diff --git a/utils/phmt.cc b/utils/phmt.cc
new file mode 100644
index 00000000..48d9f093
--- /dev/null
+++ b/utils/phmt.cc
@@ -0,0 +1,40 @@
+#include "config.h"
+
+#ifndef HAVE_CMPH
+int main() {
+ return 0;
+}
+#else
+
+#include <iostream>
+#include "weights.h"
+#include "fdict.h"
+
+using namespace std;
+
+int main(int argc, char** argv) {
+ if (argc != 2) { cerr << "Usage: " << argv[0] << " file.mphf\n"; return 1; }
+ FD::EnableHash(argv[1]);
+ cerr << "Number of keys: " << FD::NumFeats() << endl;
+ cerr << "LexFE = " << FD::Convert("LexFE") << endl;
+ cerr << "LexEF = " << FD::Convert("LexEF") << endl;
+ {
+ vector<weight_t> v(FD::NumFeats());
+ v[FD::Convert("LexFE")] = 1.0;
+ v[FD::Convert("LexEF")] = 0.5;
+ cerr << "Writing...\n";
+ Weights::WriteToFile("weights.bin", v);
+ cerr << "Done.\n";
+ }
+ {
+ vector<weight_t> v(FD::NumFeats());
+ cerr << "Reading...\n";
+ Weights::InitFromFile("weights.bin", &v);
+ cerr << "Done.\n";
+ assert(v[FD::Convert("LexFE")] == 1.0);
+ assert(v[FD::Convert("LexEF")] == 0.5);
+ }
+}
+
+#endif
+
diff --git a/utils/reconstruct_weights.cc b/utils/reconstruct_weights.cc
new file mode 100644
index 00000000..d32e4f67
--- /dev/null
+++ b/utils/reconstruct_weights.cc
@@ -0,0 +1,68 @@
+#include <iostream>
+#include <vector>
+#include <cassert>
+
+#include <boost/program_options.hpp>
+#include <boost/program_options/variables_map.hpp>
+
+#include "filelib.h"
+#include "fdict.h"
+#include "weights.h"
+
+using namespace std;
+namespace po = boost::program_options;
+
+bool InitCommandLine(int argc, char** argv, po::variables_map* conf) {
+ po::options_description opts("Configuration options");
+ opts.add_options()
+ ("weights,w",po::value<string>(),"Input feature weights file")
+ ("keys,k",po::value<string>(),"Keys file (list of features with dummy value at start)")
+ ("cmph_perfect_hash_file,h",po::value<string>(),"cmph perfect hash function file");
+ po::options_description clo("Command line options");
+ clo.add_options()
+ ("config", po::value<string>(), "Configuration file")
+ ("help,?", "Print this help message and exit");
+ po::options_description dconfig_options, dcmdline_options;
+ dconfig_options.add(opts);
+ dcmdline_options.add(opts).add(clo);
+
+ po::store(parse_command_line(argc, argv, dcmdline_options), *conf);
+ if (conf->count("config")) {
+ ifstream config((*conf)["config"].as<string>().c_str());
+ po::store(po::parse_config_file(config, dconfig_options), *conf);
+ }
+ po::notify(*conf);
+
+ if (conf->count("help") || !conf->count("cmph_perfect_hash_file") || !conf->count("weights") || !conf->count("keys")) {
+ cerr << "Generate a text format weights file. Options -w -k and -h are required.\n";
+ cerr << dcmdline_options << endl;
+ return false;
+ }
+ return true;
+}
+
+int main(int argc, char** argv) {
+ po::variables_map conf;
+ if (!InitCommandLine(argc, argv, &conf))
+ return false;
+
+ FD::EnableHash(conf["cmph_perfect_hash_file"].as<string>());
+
+ // load weights
+ vector<weight_t> weights;
+ Weights::InitFromFile(conf["weights"].as<string>(), &weights);
+
+ ReadFile rf(conf["keys"].as<string>());
+ istream& in = *rf.stream();
+ string key;
+ size_t lc = 0;
+ while(getline(in, key)) {
+ ++lc;
+ if (lc == 1) continue;
+ assert(lc <= weights.size());
+ cout << key << " " << weights[lc - 1] << endl;
+ }
+
+ return 0;
+}
+
diff --git a/utils/sampler.h b/utils/sampler.h
index 8567e922..cae660d2 100644
--- a/utils/sampler.h
+++ b/utils/sampler.h
@@ -105,7 +105,7 @@ class SampleSet {
const F& operator[](int i) const { return m_scores[i]; }
F& operator[](int i) { return m_scores[i]; }
bool empty() const { return m_scores.empty(); }
- void add(const prob_t& s) { m_scores.push_back(s); }
+ void add(const F& s) { m_scores.push_back(s); }
void clear() { m_scores.clear(); }
size_t size() const { return m_scores.size(); }
void resize(int size) { m_scores.resize(size); }
diff --git a/utils/sparse_vector.h b/utils/sparse_vector.h
index a55436fb..049151f7 100644
--- a/utils/sparse_vector.h
+++ b/utils/sparse_vector.h
@@ -1,44 +1,6 @@
#ifndef _SPARSE_VECTOR_H_
#define _SPARSE_VECTOR_H_
-#if 0
-
-#if HAVE_BOOST_ARCHIVE_TEXT_OARCHIVE_HPP
- friend class boost::serialization::access;
- template<class Archive>
- void save(Archive & ar, const unsigned int version) const {
- (void) version;
- int eff_size = values_.size();
- const_iterator it = this->begin();
- if (values_.find(0) != values_.end()) { ++it; --eff_size; }
- ar & eff_size;
- while (it != this->end()) {
- const std::pair<const std::string&, const T&> wire_pair(FD::Convert(it->first), it->second);
- ar & wire_pair;
- ++it;
- }
- }
- template<class Archive>
- void load(Archive & ar, const unsigned int version) {
- (void) version;
- this->clear();
- int sz; ar & sz;
- for (int i = 0; i < sz; ++i) {
- std::pair<std::string, T> wire_pair;
- ar & wire_pair;
- this->set_value(FD::Convert(wire_pair.first), wire_pair.second);
- }
- }
- BOOST_SERIALIZATION_SPLIT_MEMBER()
-#endif
-};
-
-#if HAVE_BOOST_ARCHIVE_TEXT_OARCHIVE_HPP
-BOOST_CLASS_TRACKING(SparseVector<double>,track_never)
-#endif
-
-#endif /// FIX
-
#include "fast_sparse_vector.h"
#define SparseVector FastSparseVector
diff --git a/utils/stringlib.cc b/utils/stringlib.cc
index 7aaee9f0..1a152985 100644
--- a/utils/stringlib.cc
+++ b/utils/stringlib.cc
@@ -2,6 +2,7 @@
#include <cstring>
#include <cstdlib>
+#include <cstdio>
#include <cassert>
#include <iostream>
#include <map>
@@ -32,7 +33,12 @@ void ParseTranslatorInput(const string& line, string* input, string* ref) {
void ProcessAndStripSGML(string* pline, map<string, string>* out) {
map<string, string>& meta = *out;
string& line = *pline;
- string lline = LowercaseString(line);
+ string lline = *pline;
+ if (lline.find("<SEG")==0 || lline.find("<Seg")==0) {
+ cerr << "Segment tags <seg> must be lowercase!\n";
+ cerr << " " << *pline << endl;
+ abort();
+ }
if (lline.find("<seg")!=0) return;
size_t close = lline.find(">");
if (close == string::npos) return; // error
@@ -85,3 +91,365 @@ void ProcessAndStripSGML(string* pline, map<string, string>* out) {
}
}
+string SGMLOpenSegTag(const map<string, string>& attr) {
+ ostringstream os;
+ os << "<seg";
+ for (map<string,string>::const_iterator it = attr.begin(); it != attr.end(); ++it)
+ os << ' ' << it->first << '=' << '"' << it->second << '"';
+ os << '>';
+ return os.str();
+}
+
+class MD5 {
+public:
+ typedef unsigned int size_type; // must be 32bit
+
+ MD5();
+ MD5(const string& text);
+ void update(const unsigned char *buf, size_type length);
+ void update(const char *buf, size_type length);
+ MD5& finalize();
+ string hexdigest() const;
+
+private:
+ void init();
+ typedef unsigned char uint1; // 8bit
+ typedef unsigned int uint4; // 32bit
+ enum {blocksize = 64}; // VC6 won't eat a const static int here
+
+ void transform(const uint1 block[blocksize]);
+ static void decode(uint4 output[], const uint1 input[], size_type len);
+ static void encode(uint1 output[], const uint4 input[], size_type len);
+
+ bool finalized;
+ uint1 buffer[blocksize]; // bytes that didn't fit in last 64 byte chunk
+ uint4 count[2]; // 64bit counter for number of bits (lo, hi)
+ uint4 state[4]; // digest so far
+ uint1 digest[16]; // the result
+
+ // low level logic operations
+ static inline uint4 F(uint4 x, uint4 y, uint4 z);
+ static inline uint4 G(uint4 x, uint4 y, uint4 z);
+ static inline uint4 H(uint4 x, uint4 y, uint4 z);
+ static inline uint4 I(uint4 x, uint4 y, uint4 z);
+ static inline uint4 rotate_left(uint4 x, int n);
+ static inline void FF(uint4 &a, uint4 b, uint4 c, uint4 d, uint4 x, uint4 s, uint4 ac);
+ static inline void GG(uint4 &a, uint4 b, uint4 c, uint4 d, uint4 x, uint4 s, uint4 ac);
+ static inline void HH(uint4 &a, uint4 b, uint4 c, uint4 d, uint4 x, uint4 s, uint4 ac);
+ static inline void II(uint4 &a, uint4 b, uint4 c, uint4 d, uint4 x, uint4 s, uint4 ac);
+};
+
+// Constants for MD5Transform routine.
+#define S11 7
+#define S12 12
+#define S13 17
+#define S14 22
+#define S21 5
+#define S22 9
+#define S23 14
+#define S24 20
+#define S31 4
+#define S32 11
+#define S33 16
+#define S34 23
+#define S41 6
+#define S42 10
+#define S43 15
+#define S44 21
+
+///////////////////////////////////////////////
+
+// F, G, H and I are basic MD5 functions.
+inline MD5::uint4 MD5::F(uint4 x, uint4 y, uint4 z) {
+ return (x&y) | (~x&z);
+}
+
+inline MD5::uint4 MD5::G(uint4 x, uint4 y, uint4 z) {
+ return (x&z) | (y&~z);
+}
+
+inline MD5::uint4 MD5::H(uint4 x, uint4 y, uint4 z) {
+ return x^y^z;
+}
+
+inline MD5::uint4 MD5::I(uint4 x, uint4 y, uint4 z) {
+ return y ^ (x | ~z);
+}
+
+// rotate_left rotates x left n bits.
+inline MD5::uint4 MD5::rotate_left(uint4 x, int n) {
+ return (x << n) | (x >> (32-n));
+}
+
+// FF, GG, HH, and II transformations for rounds 1, 2, 3, and 4.
+// Rotation is separate from addition to prevent recomputation.
+inline void MD5::FF(uint4 &a, uint4 b, uint4 c, uint4 d, uint4 x, uint4 s, uint4 ac) {
+ a = rotate_left(a+ F(b,c,d) + x + ac, s) + b;
+}
+
+inline void MD5::GG(uint4 &a, uint4 b, uint4 c, uint4 d, uint4 x, uint4 s, uint4 ac) {
+ a = rotate_left(a + G(b,c,d) + x + ac, s) + b;
+}
+
+inline void MD5::HH(uint4 &a, uint4 b, uint4 c, uint4 d, uint4 x, uint4 s, uint4 ac) {
+ a = rotate_left(a + H(b,c,d) + x + ac, s) + b;
+}
+
+inline void MD5::II(uint4 &a, uint4 b, uint4 c, uint4 d, uint4 x, uint4 s, uint4 ac) {
+ a = rotate_left(a + I(b,c,d) + x + ac, s) + b;
+}
+
+//////////////////////////////////////////////
+
+// default ctor, just initailize
+MD5::MD5()
+{
+ init();
+}
+
+//////////////////////////////////////////////
+
+// nifty shortcut ctor, compute MD5 for string and finalize it right away
+MD5::MD5(const string &text)
+{
+ init();
+ update(text.c_str(), text.length());
+ finalize();
+}
+
+//////////////////////////////
+
+void MD5::init()
+{
+ finalized=false;
+
+ count[0] = 0;
+ count[1] = 0;
+
+ // load magic initialization constants.
+ state[0] = 0x67452301;
+ state[1] = 0xefcdab89;
+ state[2] = 0x98badcfe;
+ state[3] = 0x10325476;
+}
+
+//////////////////////////////
+
+// decodes input (unsigned char) into output (uint4). Assumes len is a multiple of 4.
+void MD5::decode(uint4 output[], const uint1 input[], size_type len)
+{
+ for (unsigned int i = 0, j = 0; j < len; i++, j += 4)
+ output[i] = ((uint4)input[j]) | (((uint4)input[j+1]) << 8) |
+ (((uint4)input[j+2]) << 16) | (((uint4)input[j+3]) << 24);
+}
+
+//////////////////////////////
+
+// encodes input (uint4) into output (unsigned char). Assumes len is
+// a multiple of 4.
+void MD5::encode(uint1 output[], const uint4 input[], size_type len)
+{
+ for (size_type i = 0, j = 0; j < len; i++, j += 4) {
+ output[j] = input[i] & 0xff;
+ output[j+1] = (input[i] >> 8) & 0xff;
+ output[j+2] = (input[i] >> 16) & 0xff;
+ output[j+3] = (input[i] >> 24) & 0xff;
+ }
+}
+
+//////////////////////////////
+
+// apply MD5 algo on a block
+void MD5::transform(const uint1 block[blocksize])
+{
+ uint4 a = state[0], b = state[1], c = state[2], d = state[3], x[16];
+ decode (x, block, blocksize);
+
+ /* Round 1 */
+ FF (a, b, c, d, x[ 0], S11, 0xd76aa478); /* 1 */
+ FF (d, a, b, c, x[ 1], S12, 0xe8c7b756); /* 2 */
+ FF (c, d, a, b, x[ 2], S13, 0x242070db); /* 3 */
+ FF (b, c, d, a, x[ 3], S14, 0xc1bdceee); /* 4 */
+ FF (a, b, c, d, x[ 4], S11, 0xf57c0faf); /* 5 */
+ FF (d, a, b, c, x[ 5], S12, 0x4787c62a); /* 6 */
+ FF (c, d, a, b, x[ 6], S13, 0xa8304613); /* 7 */
+ FF (b, c, d, a, x[ 7], S14, 0xfd469501); /* 8 */
+ FF (a, b, c, d, x[ 8], S11, 0x698098d8); /* 9 */
+ FF (d, a, b, c, x[ 9], S12, 0x8b44f7af); /* 10 */
+ FF (c, d, a, b, x[10], S13, 0xffff5bb1); /* 11 */
+ FF (b, c, d, a, x[11], S14, 0x895cd7be); /* 12 */
+ FF (a, b, c, d, x[12], S11, 0x6b901122); /* 13 */
+ FF (d, a, b, c, x[13], S12, 0xfd987193); /* 14 */
+ FF (c, d, a, b, x[14], S13, 0xa679438e); /* 15 */
+ FF (b, c, d, a, x[15], S14, 0x49b40821); /* 16 */
+
+ /* Round 2 */
+ GG (a, b, c, d, x[ 1], S21, 0xf61e2562); /* 17 */
+ GG (d, a, b, c, x[ 6], S22, 0xc040b340); /* 18 */
+ GG (c, d, a, b, x[11], S23, 0x265e5a51); /* 19 */
+ GG (b, c, d, a, x[ 0], S24, 0xe9b6c7aa); /* 20 */
+ GG (a, b, c, d, x[ 5], S21, 0xd62f105d); /* 21 */
+ GG (d, a, b, c, x[10], S22, 0x2441453); /* 22 */
+ GG (c, d, a, b, x[15], S23, 0xd8a1e681); /* 23 */
+ GG (b, c, d, a, x[ 4], S24, 0xe7d3fbc8); /* 24 */
+ GG (a, b, c, d, x[ 9], S21, 0x21e1cde6); /* 25 */
+ GG (d, a, b, c, x[14], S22, 0xc33707d6); /* 26 */
+ GG (c, d, a, b, x[ 3], S23, 0xf4d50d87); /* 27 */
+ GG (b, c, d, a, x[ 8], S24, 0x455a14ed); /* 28 */
+ GG (a, b, c, d, x[13], S21, 0xa9e3e905); /* 29 */
+ GG (d, a, b, c, x[ 2], S22, 0xfcefa3f8); /* 30 */
+ GG (c, d, a, b, x[ 7], S23, 0x676f02d9); /* 31 */
+ GG (b, c, d, a, x[12], S24, 0x8d2a4c8a); /* 32 */
+
+ /* Round 3 */
+ HH (a, b, c, d, x[ 5], S31, 0xfffa3942); /* 33 */
+ HH (d, a, b, c, x[ 8], S32, 0x8771f681); /* 34 */
+ HH (c, d, a, b, x[11], S33, 0x6d9d6122); /* 35 */
+ HH (b, c, d, a, x[14], S34, 0xfde5380c); /* 36 */
+ HH (a, b, c, d, x[ 1], S31, 0xa4beea44); /* 37 */
+ HH (d, a, b, c, x[ 4], S32, 0x4bdecfa9); /* 38 */
+ HH (c, d, a, b, x[ 7], S33, 0xf6bb4b60); /* 39 */
+ HH (b, c, d, a, x[10], S34, 0xbebfbc70); /* 40 */
+ HH (a, b, c, d, x[13], S31, 0x289b7ec6); /* 41 */
+ HH (d, a, b, c, x[ 0], S32, 0xeaa127fa); /* 42 */
+ HH (c, d, a, b, x[ 3], S33, 0xd4ef3085); /* 43 */
+ HH (b, c, d, a, x[ 6], S34, 0x4881d05); /* 44 */
+ HH (a, b, c, d, x[ 9], S31, 0xd9d4d039); /* 45 */
+ HH (d, a, b, c, x[12], S32, 0xe6db99e5); /* 46 */
+ HH (c, d, a, b, x[15], S33, 0x1fa27cf8); /* 47 */
+ HH (b, c, d, a, x[ 2], S34, 0xc4ac5665); /* 48 */
+
+ /* Round 4 */
+ II (a, b, c, d, x[ 0], S41, 0xf4292244); /* 49 */
+ II (d, a, b, c, x[ 7], S42, 0x432aff97); /* 50 */
+ II (c, d, a, b, x[14], S43, 0xab9423a7); /* 51 */
+ II (b, c, d, a, x[ 5], S44, 0xfc93a039); /* 52 */
+ II (a, b, c, d, x[12], S41, 0x655b59c3); /* 53 */
+ II (d, a, b, c, x[ 3], S42, 0x8f0ccc92); /* 54 */
+ II (c, d, a, b, x[10], S43, 0xffeff47d); /* 55 */
+ II (b, c, d, a, x[ 1], S44, 0x85845dd1); /* 56 */
+ II (a, b, c, d, x[ 8], S41, 0x6fa87e4f); /* 57 */
+ II (d, a, b, c, x[15], S42, 0xfe2ce6e0); /* 58 */
+ II (c, d, a, b, x[ 6], S43, 0xa3014314); /* 59 */
+ II (b, c, d, a, x[13], S44, 0x4e0811a1); /* 60 */
+ II (a, b, c, d, x[ 4], S41, 0xf7537e82); /* 61 */
+ II (d, a, b, c, x[11], S42, 0xbd3af235); /* 62 */
+ II (c, d, a, b, x[ 2], S43, 0x2ad7d2bb); /* 63 */
+ II (b, c, d, a, x[ 9], S44, 0xeb86d391); /* 64 */
+
+ state[0] += a;
+ state[1] += b;
+ state[2] += c;
+ state[3] += d;
+
+ // Zeroize sensitive information.
+ memset(x, 0, sizeof x);
+}
+
+//////////////////////////////
+
+// MD5 block update operation. Continues an MD5 message-digest
+// operation, processing another message block
+void MD5::update(const unsigned char input[], size_type length)
+{
+ // compute number of bytes mod 64
+ size_type index = count[0] / 8 % blocksize;
+
+ // Update number of bits
+ if ((count[0] += (length << 3)) < (length << 3))
+ count[1]++;
+ count[1] += (length >> 29);
+
+ // number of bytes we need to fill in buffer
+ size_type firstpart = 64 - index;
+
+ size_type i;
+
+ // transform as many times as possible.
+ if (length >= firstpart)
+ {
+ // fill buffer first, transform
+ memcpy(&buffer[index], input, firstpart);
+ transform(buffer);
+
+ // transform chunks of blocksize (64 bytes)
+ for (i = firstpart; i + blocksize <= length; i += blocksize)
+ transform(&input[i]);
+
+ index = 0;
+ }
+ else
+ i = 0;
+
+ // buffer remaining input
+ memcpy(&buffer[index], &input[i], length-i);
+}
+
+//////////////////////////////
+
+// for convenience provide a verson with signed char
+void MD5::update(const char input[], size_type length)
+{
+ update((const unsigned char*)input, length);
+}
+
+//////////////////////////////
+
+// MD5 finalization. Ends an MD5 message-digest operation, writing the
+// the message digest and zeroizing the context.
+MD5& MD5::finalize()
+{
+ static unsigned char padding[64] = {
+ 0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
+ };
+
+ if (!finalized) {
+ // Save number of bits
+ unsigned char bits[8];
+ encode(bits, count, 8);
+
+ // pad out to 56 mod 64.
+ size_type index = count[0] / 8 % 64;
+ size_type padLen = (index < 56) ? (56 - index) : (120 - index);
+ update(padding, padLen);
+
+ // Append length (before padding)
+ update(bits, 8);
+
+ // Store state in digest
+ encode(digest, state, 16);
+
+ // Zeroize sensitive information.
+ memset(buffer, 0, sizeof buffer);
+ memset(count, 0, sizeof count);
+
+ finalized=true;
+ }
+
+ return *this;
+}
+
+//////////////////////////////
+
+// return hex representation of digest as string
+string MD5::hexdigest() const {
+ if (!finalized)
+ return "";
+
+ char buf[33];
+ for (int i=0; i<16; i++)
+ sprintf(buf+i*2, "%02x", digest[i]);
+ buf[32]=0;
+
+ return string(buf);
+}
+
+//////////////////////////////
+
+string md5(const string& str) {
+ MD5 md5 = MD5(str);
+ return md5.hexdigest();
+}
+
diff --git a/utils/stringlib.h b/utils/stringlib.h
index 8022bb88..cafbdac3 100644
--- a/utils/stringlib.h
+++ b/utils/stringlib.h
@@ -249,6 +249,7 @@ inline void SplitCommandAndParam(const std::string& in, std::string* cmd, std::s
}
void ProcessAndStripSGML(std::string* line, std::map<std::string, std::string>* out);
+std::string SGMLOpenSegTag(const std::map<std::string, std::string>& attr);
// given the first character of a UTF8 block, find out how wide it is
// see http://en.wikipedia.org/wiki/UTF-8 for more info
@@ -260,4 +261,6 @@ inline unsigned int UTF8Len(unsigned char x) {
else return 0;
}
+std::string md5(const std::string& in);
+
#endif
diff --git a/utils/tdict.cc b/utils/tdict.cc
index c21b2b48..de234323 100644
--- a/utils/tdict.cc
+++ b/utils/tdict.cc
@@ -13,6 +13,10 @@ using namespace std;
Dict TD::dict_;
+unsigned int TD::NumWords() {
+ return dict_.max();
+}
+
WordID TD::Convert(const std::string& s) {
return dict_.Convert(s);
}
diff --git a/utils/ts.cc b/utils/ts.cc
index 3694e076..bf4f8f69 100644
--- a/utils/ts.cc
+++ b/utils/ts.cc
@@ -7,6 +7,7 @@
#include "prob.h"
#include "sparse_vector.h"
#include "fast_sparse_vector.h"
+#include "stringlib.h"
using namespace std;
@@ -79,6 +80,11 @@ int main() {
y -= y;
}
cerr << "Counted " << c << " times\n";
+
+ cerr << md5("this is a test") << endl;
+ cerr << md5("some other ||| string is") << endl;
+ map<string,string> x; x["id"] = "12"; x["grammar"] = "/path/to/grammar.gz";
+ cerr << SGMLOpenSegTag(x) << endl;
return 0;
}
diff --git a/utils/weights.cc b/utils/weights.cc
index 6b7e58ed..f1406cbf 100644
--- a/utils/weights.cc
+++ b/utils/weights.cc
@@ -8,101 +8,149 @@
using namespace std;
-void Weights::InitFromFile(const std::string& filename, vector<string>* feature_list) {
+void Weights::InitFromFile(const string& filename,
+ vector<weight_t>* pweights,
+ vector<string>* feature_list) {
+ vector<weight_t>& weights = *pweights;
if (!SILENT) cerr << "Reading weights from " << filename << endl;
ReadFile in_file(filename);
istream& in = *in_file.stream();
assert(in);
- int weight_count = 0;
- bool fl = false;
- string buf;
- double val = 0;
- while (in) {
- getline(in, buf);
- if (buf.size() == 0) continue;
- if (buf[0] == '#') continue;
- for (int i = 0; i < buf.size(); ++i)
- if (buf[i] == '=') buf[i] = ' ';
- int start = 0;
- while(start < buf.size() && buf[start] == ' ') ++start;
- int end = 0;
- while(end < buf.size() && buf[end] != ' ') ++end;
- const int fid = FD::Convert(buf.substr(start, end - start));
- while(end < buf.size() && buf[end] == ' ') ++end;
- val = strtod(&buf.c_str()[end], NULL);
- if (isnan(val)) {
- cerr << FD::Convert(fid) << " has weight NaN!\n";
- abort();
+
+ bool read_text = true;
+ if (1) {
+ ReadFile hdrrf(filename);
+ istream& hi = *hdrrf.stream();
+ assert(hi);
+ char buf[10];
+ hi.read(buf, 5);
+ assert(hi.good());
+ if (strncmp(buf, "_PHWf", 5) == 0) {
+ read_text = false;
+ }
+ }
+
+ if (read_text) {
+ int weight_count = 0;
+ bool fl = false;
+ string buf;
+ weight_t val = 0;
+ while (in) {
+ getline(in, buf);
+ if (buf.size() == 0) continue;
+ if (buf[0] == '#') continue;
+ if (buf[0] == ' ') {
+ cerr << "Weights file lines may not start with whitespace.\n" << buf << endl;
+ abort();
+ }
+ for (int i = buf.size() - 1; i > 0; --i)
+ if (buf[i] == '=' || buf[i] == '\t') { buf[i] = ' '; break; }
+ int start = 0;
+ while(start < buf.size() && buf[start] == ' ') ++start;
+ int end = 0;
+ while(end < buf.size() && buf[end] != ' ') ++end;
+ const int fid = FD::Convert(buf.substr(start, end - start));
+ if (feature_list) { feature_list->push_back(buf.substr(start, end - start)); }
+ while(end < buf.size() && buf[end] == ' ') ++end;
+ val = strtod(&buf.c_str()[end], NULL);
+ if (isnan(val)) {
+ cerr << FD::Convert(fid) << " has weight NaN!\n";
+ abort();
+ }
+ if (weights.size() <= fid)
+ weights.resize(fid + 1);
+ weights[fid] = val;
+ ++weight_count;
+ if (!SILENT) {
+ if (weight_count % 50000 == 0) { cerr << '.' << flush; fl = true; }
+ if (weight_count % 2000000 == 0) { cerr << " [" << weight_count << "]\n"; fl = false; }
+ }
}
- if (wv_.size() <= fid)
- wv_.resize(fid + 1);
- wv_[fid] = val;
- if (feature_list) { feature_list->push_back(FD::Convert(fid)); }
- ++weight_count;
if (!SILENT) {
- if (weight_count % 50000 == 0) { cerr << '.' << flush; fl = true; }
- if (weight_count % 2000000 == 0) { cerr << " [" << weight_count << "]\n"; fl = false; }
+ if (fl) { cerr << endl; }
+ cerr << "Loaded " << weight_count << " feature weights\n";
+ }
+ } else { // !read_text
+ char buf[6];
+ in.read(buf, 5);
+ size_t num_keys;
+ in.read(reinterpret_cast<char*>(&num_keys), sizeof(size_t));
+ if (num_keys != FD::NumFeats()) {
+ cerr << "Hash function reports " << FD::NumFeats() << " keys but weights file contains " << num_keys << endl;
+ abort();
+ }
+ weights.resize(num_keys);
+ in.read(reinterpret_cast<char*>(&weights.front()), num_keys * sizeof(weight_t));
+ if (!in.good()) {
+ cerr << "Error loading weights!\n";
+ abort();
+ } else {
+ cerr << " Successfully loaded " << (num_keys * sizeof(weight_t)) << " bytes\n";
}
- }
- if (!SILENT) {
- if (fl) { cerr << endl; }
- cerr << "Loaded " << weight_count << " feature weights\n";
}
}
-void Weights::WriteToFile(const std::string& fname, bool hide_zero_value_features, const string* extra) const {
+void Weights::WriteToFile(const string& fname,
+ const vector<weight_t>& weights,
+ bool hide_zero_value_features,
+ const string* extra) {
WriteFile out(fname);
ostream& o = *out.stream();
assert(o);
- if (extra) { o << "# " << *extra << endl; }
- o.precision(17);
- const int num_feats = FD::NumFeats();
- for (int i = 1; i < num_feats; ++i) {
- const double val = (i < wv_.size() ? wv_[i] : 0.0);
- if (hide_zero_value_features && val == 0.0) continue;
- o << FD::Convert(i) << ' ' << val << endl;
- }
-}
-
-void Weights::InitVector(std::vector<double>* w) const {
- *w = wv_;
-}
+ bool write_text = !FD::UsingPerfectHashFunction();
-void Weights::InitSparseVector(SparseVector<double>* w) const {
- for (int i = 1; i < wv_.size(); ++i) {
- const double& weight = wv_[i];
- if (weight) w->set_value(i, weight);
+ if (write_text) {
+ if (extra) { o << "# " << *extra << endl; }
+ o.precision(17);
+ const int num_feats = FD::NumFeats();
+ for (int i = 1; i < num_feats; ++i) {
+ const weight_t val = (i < weights.size() ? weights[i] : 0.0);
+ if (hide_zero_value_features && val == 0.0) continue;
+ o << FD::Convert(i) << ' ' << val << endl;
+ }
+ } else {
+ o.write("_PHWf", 5);
+ const size_t keys = FD::NumFeats();
+ assert(keys <= weights.size());
+ o.write(reinterpret_cast<const char*>(&keys), sizeof(keys));
+ o.write(reinterpret_cast<const char*>(&weights[0]), keys * sizeof(weight_t));
}
}
-void Weights::InitFromVector(const std::vector<double>& w) {
- wv_ = w;
- if (wv_.size() > FD::NumFeats())
- cerr << "WARNING: initializing weight vector has more features than the global feature dictionary!\n";
- wv_.resize(FD::NumFeats(), 0);
-}
-
-void Weights::InitFromVector(const SparseVector<double>& w) {
- wv_.clear();
- wv_.resize(FD::NumFeats(), 0.0);
- for (int i = 1; i < FD::NumFeats(); ++i)
- wv_[i] = w.value(i);
+void Weights::InitSparseVector(const vector<weight_t>& dv,
+ SparseVector<weight_t>* sv) {
+ sv->clear();
+ for (unsigned i = 1; i < dv.size(); ++i) {
+ if (dv[i]) sv->set_value(i, dv[i]);
+ }
}
-void Weights::SetWeight(SparseVector<double>* v, const string fname, const double w) {
- WordID fid = FD::Convert(fname);
- cout << "fid " << fid << endl;
- SetWeight(v, fid, w);
+void Weights::SanityCheck(const vector<weight_t>& w) {
+ for (int i = 0; i < w.size(); ++i) {
+ assert(!isnan(w[i]));
+ assert(!isinf(w[i]));
+ }
}
-void Weights::SetWeight(SparseVector<double>* v, const WordID fid, const double w) {
- wv_.resize(FD::NumFeats(), 0.0);
- wv_[fid] = w;
- //v->set_value(fid, w);
-}
+struct FComp {
+ const vector<weight_t>& w_;
+ FComp(const vector<weight_t>& w) : w_(w) {}
+ bool operator()(int a, int b) const {
+ return fabs(w_[a]) > fabs(w_[b]);
+ }
+};
-void Weights::sz()
-{
- cout << "wv_.size() " << wv_.size() << endl;
+void Weights::ShowLargestFeatures(const vector<weight_t>& w) {
+ vector<int> fnums(w.size());
+ for (int i = 0; i < w.size(); ++i)
+ fnums[i] = i;
+ vector<int>::iterator mid = fnums.begin();
+ mid += (w.size() > 10 ? 10 : w.size());
+ partial_sort(fnums.begin(), mid, fnums.end(), FComp(w));
+ cerr << "TOP FEATURES:";
+ for (vector<int>::iterator i = fnums.begin(); i != mid; ++i) {
+ cerr << ' ' << FD::Convert(*i) << '=' << w[*i];
+ }
+ cerr << endl;
}
diff --git a/utils/weights.h b/utils/weights.h
index 86701add..30f71db0 100644
--- a/utils/weights.h
+++ b/utils/weights.h
@@ -2,25 +2,29 @@
#define _WEIGHTS_H_
#include <string>
-#include <map>
#include <vector>
#include "sparse_vector.h"
+// warning: in the future this will become float
+typedef double weight_t;
+
class Weights {
public:
- Weights() {}
- void InitFromFile(const std::string& fname, std::vector<std::string>* feature_list = NULL);
- void WriteToFile(const std::string& fname, bool hide_zero_value_features = true, const std::string* extra = NULL) const;
- void InitVector(std::vector<double>* w) const;
- void InitSparseVector(SparseVector<double>* w) const;
- void InitFromVector(const std::vector<double>& w);
- void InitFromVector(const SparseVector<double>& w);
- void SetWeight(SparseVector<double>* v, const std::string f, const double w);
- void SetWeight(SparseVector<double>* v, const WordID fid, const double w);
- std::vector<double>* getw() { return &wv_; }; // probably a hack
- void sz();
+ static void InitFromFile(const std::string& fname,
+ std::vector<weight_t>* weights,
+ std::vector<std::string>* feature_list = NULL);
+ static void WriteToFile(const std::string& fname,
+ const std::vector<weight_t>& weights,
+ bool hide_zero_value_features = true,
+ const std::string* extra = NULL);
+ static void InitSparseVector(const std::vector<weight_t>& dv,
+ SparseVector<weight_t>* sv);
+ // check for infinities, NaNs, etc
+ static void SanityCheck(const std::vector<weight_t>& w);
+ // write weights with largest magnitude to cerr
+ static void ShowLargestFeatures(const std::vector<weight_t>& w);
private:
- std::vector<double> wv_;
+ Weights();
};
#endif
diff --git a/utils/weights_test.cc b/utils/weights_test.cc
index 8a4c26ef..938b311f 100644
--- a/utils/weights_test.cc
+++ b/utils/weights_test.cc
@@ -14,11 +14,10 @@ class WeightsTest : public testing::Test {
virtual void TearDown() { }
};
-
TEST_F(WeightsTest,Load) {
- Weights w;
- w.InitFromFile("test_data/weights");
- w.WriteToFile("-");
+ vector<weight_t> v;
+ Weights::InitFromFile("test_data/weights", &v);
+ Weights::WriteToFile("-", v);
}
int main(int argc, char **argv) {