pretty

git-svn-id: https://ws10smt.googlecode.com/svn/trunk@390 ec762483-ff6d-05da-a07a-a48fb63a330f

4 files changed, 35 insertions, 13 deletions

diff --git a/decoder/ff_fsa.h b/decoder/ff_fsa.h
index 8ca1951f..41343f6c 100755
--- a/decoder/ff_fsa.h
+++ b/decoder/ff_fsa.h
@@ -62,8 +62,8 @@ protected:
     std::memcpy(state,begin,n);
   }
   template <class T>
-  inline void static to_state(void *state,T const* begin,int n) {
-    to_state(state,(char const*)begin,n);
+  inline void static to_state(void *state,T const* begin,int n=1) {
+    to_state(state,(char const*)begin,n*sizeof(T));
   }
   template <class T>
   inline void static to_state(void *state,T const* begin,T const* end) {
diff --git a/decoder/ff_sample_fsa.h b/decoder/ff_sample_fsa.h
index 8befc0bb..d8aa7830 100755
--- a/decoder/ff_sample_fsa.h
+++ b/decoder/ff_sample_fsa.h
@@ -31,11 +31,12 @@ typedef FeatureFunctionFromFsa<WordPenaltyFsa> WordPenaltyFromFsa;
 
 //
 struct LongerThanPrev : public FsaFeatureFunctionBase<LongerThanPrev> {
+  typedef FsaFeatureFunctionBase<LongerThanPrev> Base;
   static std::string usage(bool param,bool verbose) {
     return FeatureFunction::usage_helper(
       "LongerThanPrev",
       "",
-      "stupid example stateful (bigram) feature: -1 per target word that's longer than the previous word (always fires for first word of sentence)",
+      "stupid example stateful (bigram) feature: -1 per target word that's longer than the previous word (<s> sentence begin considered 3 chars long, </s> is sentence end.)",
       param,verbose);
   }
 
@@ -49,16 +50,21 @@ struct LongerThanPrev : public FsaFeatureFunctionBase<LongerThanPrev> {
     return std::strlen(TD::Convert(w));
   }
   int markov_order() const { return 1; }
-  LongerThanPrev(std::string const& param) {
+  LongerThanPrev(std::string const& param) : Base(sizeof(int),singleton_sentence(TD::se)) {
     Init();
-    set_state_bytes(sizeof(int));
-//    start.resize(state_bytes()); // this is done by set_state_bytes already.
-//    h_start.resize(state_bytes());
-//    int ss=-1;
-//    wordcpy((WordID*)start.begin(),&ss,&ss+1);
-    //to_state(start.begin(),&ss,1);
-    wordlen(start.begin())=-1;
+    if (0) { // all this is done in constructor already
+      set_state_bytes(sizeof(int));
+      start.resize(state_bytes()); // this is done by set_state_bytes already.
+      h_start.resize(state_bytes());
+      int ss=3;
+      to_state(start.begin(),&ss,1);
+      ss=4;
+      to_state(h_start.begin(),&ss,1);
+    }
+
+    wordlen(start.begin())=3;
     wordlen(h_start.begin())=4; // estimate: anything >4 chars is usually longer than previous
+
   }
 
   static const float val_per_target_word=-1;
@@ -79,7 +85,7 @@ struct ShorterThanPrev : FsaTypedBase<int,ShorterThanPrev> {
     return FeatureFunction::usage_helper(
       "ShorterThanPrev",
       "",
-      "stupid example stateful (bigram) feature: -1 per target word that's shorter than the previous word (always fires for end of sentence)",
+      "stupid example stateful (bigram) feature: -1 per target word that's shorter than the previous word (end of sentence considered '</s>')",
       param,verbose);
   }
 
@@ -87,7 +93,7 @@ struct ShorterThanPrev : FsaTypedBase<int,ShorterThanPrev> {
     return std::strlen(TD::Convert(w));
   }
   ShorterThanPrev(std::string const& param)
-  : Base(-1,4,Sentence(1,TD::Convert("")))
+  : Base(3,4,singleton_sentence(TD::se))
     // start, h_start, end_phrase
     // estimate: anything <4 chars is usually shorter than previous
   {
diff --git a/decoder/filelib.h b/decoder/filelib.h
index 2d347c00..af66dd05 100644
--- a/decoder/filelib.h
+++ b/decoder/filelib.h
@@ -6,6 +6,7 @@
 #include <iostream>
 #include <cstdlib>
 #include <boost/shared_ptr.hpp>
+#include <stdexcept>
 #include "gzstream.h"
 
 bool FileExists(const std::string& file_name);
@@ -37,6 +38,10 @@ struct BaseFile {
   }
   std::string filename_;
 protected:
+  void error(std::string const& reason,std::string const& filename) {
+    throw std::runtime_error("File "+filename+" - "+reason);
+  }
+
   PS ps_;
   static bool EndsWith(const std::string& f, const std::string& suf) {
     return (f.size() > suf.size()) && (f.rfind(suf) == f.size() - suf.size());
@@ -56,6 +61,7 @@ class ReadFile : public BaseFile<std::istream> {
     } else {
       if (!FileExists(filename)) {
         std::cerr << "File does not exist: " << filename << std::endl;
+        error(filename," couldn't read nonexistant file.");
         abort();
       }
       char const* file=filename_.c_str(); // just in case the gzstream keeps using the filename for longer than the constructor, e.g. inflateReset2.  warning in valgrind that I'm hoping will disappear - it makes no sense.
@@ -64,6 +70,7 @@ class ReadFile : public BaseFile<std::istream> {
              static_cast<std::istream*>(new std::ifstream(file)));
       if (!*ps_) {
         std::cerr << "Failed to open " << filename << std::endl;
+        error(filename," open for reading failed.");
         abort();
       }
     }
@@ -86,6 +93,7 @@ class WriteFile : public BaseFile<std::ostream> {
                 static_cast<std::ostream*>(new std::ofstream(file)));
       if (!*ps_) {
         std::cerr << "Failed to open " << filename << std::endl;
+        error(filename," open for writing failed.");
         abort();
       }
     }
diff --git a/decoder/sentences.h b/decoder/sentences.h
index 482d3be9..452fb21e 100755
--- a/decoder/sentences.h
+++ b/decoder/sentences.h
@@ -30,6 +30,14 @@ inline void wordcpy(WordID *dest,WordID const* src,WordID const* src_end) {
   wordcpy(dest,src,src_end-src);
 }
 
+inline Sentence singleton_sentence(WordID t) {
+  return Sentence(1,t);
+}
+
+inline Sentence singleton_sentence(std::string const& s) {
+  return singleton_sentence(TD::Convert(s));
+}
+
 
 inline std::ostream & operator<<(std::ostream &out,Sentence const& s) {
   return out<<TD::GetString(s);