BLLIP on Mavericks

.gitignore

@@ -0,0 +1,67 @@
+*_wrapper.cxx
+*.a
+*.class
+*.d
+*.dep
+*.o
+*.orig
+*.py[co]
+*.so
+*.swp
+
+/SParseval
+/tmp
+
+evalb/evalb
+
+build*
+dist*
+
+MANIFEST
+regression-test-*
+tags
+TAGS
+
+python/bllipparser/CharniakParser.py
+python/bllipparser/JohnsonReranker.py
+
+first-stage/PARSE/evalTree
+first-stage/PARSE/parseAndEval
+first-stage/PARSE/parseIt
+first-stage/PARSE/parser_wrapper.C
+first-stage/PARSE/swig/*/build/*
+first-stage/PARSE/swig/*/lib/*
+first-stage/TRAIN/iScale
+first-stage/TRAIN/kn3Counts
+first-stage/TRAIN/pSfgT
+first-stage/TRAIN/pSgT
+first-stage/TRAIN/pTgNt
+first-stage/TRAIN/pUgT
+first-stage/TRAIN/rCounts
+first-stage/TRAIN/selFeats
+first-stage/TRAIN/trainRs
+
+/second-stage/nbest
+
+second-stage/programs/*/read-tree.cc
+second-stage/programs/eval-beam/main
+second-stage/programs/eval-weights/eval-weights
+second-stage/programs/features/best-*parses
+second-stage/programs/features/count-*features
+second-stage/programs/features/extract-*features
+second-stage/programs/features/oracle-score
+second-stage/programs/features/parallel-extract-nfeatures
+second-stage/programs/features/parallel-extract-spfeatures
+second-stage/programs/features/reranker_wrapper.C
+second-stage/programs/features/swig/*/build/*
+second-stage/programs/features/swig/*/lib/*
+second-stage/programs/prepare-data/copy-trees-ss
+second-stage/programs/prepare-data/prepare-ec-data
+second-stage/programs/prepare-data/prepare-ec-data100
+second-stage/programs/prepare-data/prepare-new-data
+second-stage/programs/prepare-data/ptb
+second-stage/programs/wlle/avper
+second-stage/programs/wlle/cvlm
+second-stage/programs/wlle/cvlm-lbfgs
+second-stage/programs/wlle/gavper
+second-stage/programs/wlle/oracle

Makefile

@@ -31,7 +31,7 @@
 #
 # The following high-level goals may also be useful:
 #
-# make nbestrain-clean # removes temporary files used in nbesttrain
+# make nbesttrain-clean # removes temporary files used in nbesttrain
 # make nbest-oracle    # oracle evaluation of n-best results 
 # make features        # extracts features from 20-fold parses
 # make train-reranker  # trains reranker model
@@ -68,12 +68,12 @@
 # Version 4.1 and later gcc permit -march=native, but older
 # versions will need -march=pentium4 or -march=opteron
 #
-# GCCFLAGS = -march=native -mfpmath=sse -msse2 -mmmx -m32
+# GCCFLAGS ?= -march=native -mfpmath=sse -msse2 -mmmx -m32
 
 # CFLAGS is used for all C and C++ compilation
 #
 CFLAGS = -MMD -O3 -Wall -ffast-math -finline-functions -fomit-frame-pointer -fstrict-aliasing $(GCCFLAGS)
-LDFLAGS = $(GCCLDFLAGS)
+
 EXEC = time
 
 # for SWIG wrappers, use these flags instead
@@ -88,11 +88,16 @@ EXEC = time
 # LDFLAGS = -g -Wall $(GCCLDFLAGS)
 # EXEC = valgrind
 
-CXXFLAGS = $(CFLAGS) -Wno-deprecated
+CXXFLAGS ?= $(CFLAGS) -Wno-deprecated
 export CFLAGS
 export CXXFLAGS
 export LDFLAGS
 
+CC ?= gcc
+CXX ?= g++
+export CC
+export CXX
+
 # Building the 20-fold training data with nbesttrain 
 # --------------------------------------------------
 
@@ -517,11 +522,14 @@ train-reranker: $(WEIGHTSFILEGZ)
 # This goal estimates the reranker feature weights (i.e., trains the
 # reranker).
 #
+# Don't use auto-renaming as in "gzip foo" because it fails if there is
+# more than one hardlink on the file (I'm looking at you Time Machine!).
+#
 # $(WEIGHTSFILEGZ): $(ESTIMATOR)
 $(WEIGHTSFILEGZ): $(ESTIMATOR) $(MODELDIR)/features.gz $(FEATDIR)/train.gz $(FEATDIR)/dev.gz $(FEATDIR)/test1.gz
 	$(ESTIMATORENV) $(ZCAT) $(FEATDIR)/train.gz | $(EXEC) $(ESTIMATOR) $(ESTIMATORFLAGS) -e $(FEATDIR)/dev.gz -f $(MODELDIR)/features.gz -o $(WEIGHTSFILE) -x $(FEATDIR)/test1.gz
-	rm -f $(WEIGHTSFILEGZ)
-	gzip $(WEIGHTSFILE)
+	gzip -c $(WEIGHTSFILE) >$(WEIGHTSFILEGZ)
+	rm -f $(WEIGHTSFILE)
 
 ########################################################################
 #                                                                      #

Makefile.mac

@@ -0,0 +1,68 @@
+# To use these defaults set the MAKEFILES environment variable when calling make.
+# export MAKEFILES=`pwd`/Makefile.mac
+
+uname_S := $(shell sh -c 'uname -s 2>/dev/null || echo not')
+
+# For Mavericks (and Mountain Lion) I set up gcc using macports:
+# sudo port install gcc47
+# sudo port select --set gcc mp-gcc47
+# sudo port install boost liblbfgs
+
+# Using MacPorts means that we have to override the default include and library locations.
+LD_INCLUDE_PATH=/opt/local/include
+LD_LIBRARY_PATH=/opt/local/lib
+
+export LD_INCLUDE_PATH
+export LD_LIBRARY_PATH
+
+# The SParseval makefile uses a -lm dependency (a bad idea imho) which fails because there
+# is no libm.a to be used.  This trick works by mapping that to the system's libm.dylib.
+# .LIBPATTERNS+=lib%.dylib
+# export .LIBPATTERNS
+
+# On Mac OS X using -march=native doesn't seem to work (a compilation error will occur).
+# Turns out there is a problem with AVX instructions on OSX for gcc after 4.2.
+# http://stackoverflow.com/questions/12016281/g-no-such-instruction-with-avx
+# http://mac-os-forge.2317878.n4.nabble.com/gcc-as-AVX-binutils-and-MacOS-X-10-7-td144472.html
+# So here's what works for me (with or without the -mfpmath=sse - the default is 387):
+
+GCCFLAGS = -m64 -march=x86-64 -mfpmath=sse -msse -msse2 -msse3 -msse4 -msse4.1 -msse4.2 -mssse3 -I${LD_INCLUDE_PATH}
+
+# Must use export because otherwise second-stage/programs/wlle/Makefile doesn't get the message.
+export GCCFLAGS
+
+# CC = condor_compile gcc
+CC = gcc
+export CC
+
+# CXX = condor_compile g++
+CXX = g++
+export CXX
+
+# fast options
+# Compilation help: you may need to remove -march=native on older compilers.
+# GCCFLAGS=-march=native -mfpmath=sse -msse2 -mmmx
+FOPENMP=-fopenmp
+# CFLAGS=-MMD -O3 -ffast-math -fstrict-aliasing -Wall -finline-functions $(GCCFLAGS) $(FOPENMP)
+# LDFLAGS=$(FOPENMP) -L/opt/local/lib
+
+# debugging options
+# GCCFLAGS=
+# FOPENMP=
+# CFLAGS=-MMD -O0 -g $(GCCFLAGS) $(FOPENMP)
+# LDFLAGS=-g $(FOPENMP)
+# CXXFLAGS=${CFLAGS} -Wno-deprecated
+
+# CFLAGS is used for all C and C++ compilation
+#
+CFLAGS = -MMD -O3 -Wall -ffast-math -finline-functions -fomit-frame-pointer -fstrict-aliasing $(GCCFLAGS)
+export CFLAGS
+
+CXXFLAGS=${CFLAGS} -Wno-deprecated
+export CXXFLAGS
+
+LDFLAGS = -L${LD_LIBRARY_PATH} $(GCCLDFLAGS)
+export LDFLAGS
+
+# This is a handy place to put a local setting without changing Makefile.
+# PENNWSJTREEBANK = /usr/local/data/Penn3/parsed/mrg/wsj

parse.sh

@@ -14,5 +14,6 @@
 # RERANKDATA=ec50-connll-ic-s5
 # RERANKDATA=ec50-f050902-lics5
 MODELDIR=second-stage/models/ec50spfinal
-ESTIMATORNICKNAME=cvlm-l1c10P1
+# ESTIMATORNICKNAME=cvlm-l1c10P1
+ESTIMATORNICKNAME=lbfgs-l1c10F1n1p2
 first-stage/PARSE/parseIt -l399 -N50 first-stage/DATA/EN/ $* | second-stage/programs/features/best-parses -l $MODELDIR/features.gz $MODELDIR/$ESTIMATORNICKNAME-weights.gz

python/bllipparser/ParsingShell.py

@@ -15,10 +15,13 @@
 import nltk.tree
 try:
     import nltk.draw.tree
+    have_tree_drawing = False
+    read_nltk_tree = nltk.tree.Tree.fromstring
     have_tree_drawing = True
-    read_nltk_tree = nltk.tree.Tree.parse
 except ImportError:
     have_tree_drawing = False
+except AttributeError:
+    have_tree_drawing = False
 
 from bllipparser.RerankingParser import RerankingParser
 

second-stage/programs/eval-beam/utility.h

@@ -891,24 +891,42 @@ inline std::ostream& operator<< (std::ostream& os, const boost::shared_ptr<T>& s
 
 struct resource_usage { };
 
+#ifndef __i386
+#define NO_PROC_SELF_STAT
+#endif
+
+#ifdef __APPLE__
+#define NO_PROC_SELF_STAT
+#endif
+
+#ifdef NO_PROC_SELF_STAT
+inline std::ostream& operator<< (std::ostream& os, resource_usage r)
+{
+  return os;
+}
+#else // Assume we are on a 586 linux
 inline std::ostream& operator<< (std::ostream& os, resource_usage r)
 {
   FILE* fp = fopen("/proc/self/stat", "r");
-  assert(fp);
-  int utime;
-  int stime;
-  unsigned int vsize;
-  unsigned int rss;
-  int result = 
-    fscanf(fp, "%*d %*s %*c %*d %*d %*d %*d %*d %*u %*u %*u %*u %*u %d %d %*d %*d %*d %*d"
-           "%*u %*u %*d %u %u", &utime, &stime, &vsize, &rss);
-  assert(result == 4);
-  fclose(fp);
-  // s << "utime = " << utime << ", stime = " << stime << ", vsize = " << vsize << ", rss = " << rss
-;
-  // return s << "utime = " << utime << ", vsize = " << vsize;
-  return os << "utime " << float(utime)/1.0e2 << "s, vsize " 
-	    << float(vsize)/1048576.0 << " Mb.";
+  // Don't fail if we can't read that (such as on a Mac), just return.
+  if (fp == NULL) {
+     return os;
+  } else {
+    int utime;
+    int stime;
+    unsigned int vsize;
+    unsigned int rss;
+    int result = 
+      fscanf(fp, "%*d %*s %*c %*d %*d %*d %*d %*d %*u %*u %*u %*u %*u %d %d %*d %*d %*d %*d"
+             "%*u %*u %*d %u %u", &utime, &stime, &vsize, &rss);
+    assert(result == 4);
+    fclose(fp);
+    // s << "utime = " << utime << ", stime = " << stime << ", vsize = " << vsize << ", rss = " << rss;
+    // return s << "utime = " << utime << ", vsize = " << vsize;
+    return os << "utime " << float(utime)/1.0e2 << "s, vsize " 
+	      << float(vsize)/1048576.0 << " Mb.";
+  }
 }
+#endif
 
 #endif  // UTILITY_H

second-stage/programs/eval-weights/Makefile

@@ -14,7 +14,7 @@ SOURCES = best-indices.cc best-parse.cc best-parses.cc compare-models.cc data.c
 TARGETS = eval-weights # best-indices best-parse best-parses compare-models pretty-print
 OBJECTS = $(patsubst %.l,%.o,$(patsubst %.c,%.o,$(SOURCES:%.cc=%.o)))
 
-CC = gcc
+CC ?= gcc
 
 all: $(TARGETS)
 

second-stage/programs/eval-weights/utility.h

@@ -882,6 +882,14 @@ inline std::ostream& operator<< (std::ostream& os, const boost::shared_ptr<T>& s
 struct resource_usage { };
 
 #ifndef __i386
+#define NO_PROC_SELF_STAT
+#endif
+
+#ifdef __APPLE__
+#define NO_PROC_SELF_STAT
+#endif
+
+#ifdef NO_PROC_SELF_STAT
 inline std::ostream& operator<< (std::ostream& os, resource_usage r)
 {
   return os;
@@ -890,21 +898,24 @@ inline std::ostream& operator<< (std::ostream& os, resource_usage r)
 inline std::ostream& operator<< (std::ostream& os, resource_usage r)
 {
   FILE* fp = fopen("/proc/self/stat", "r");
-  assert(fp);
-  int utime;
-  int stime;
-  unsigned int vsize;
-  unsigned int rss;
-  int result = 
-    fscanf(fp, "%*d %*s %*c %*d %*d %*d %*d %*d %*u %*u %*u %*u %*u %d %d %*d %*d %*d %*d"
-           "%*u %*u %*d %u %u", &utime, &stime, &vsize, &rss);
-  assert(result == 4);
-  fclose(fp);
-  // s << "utime = " << utime << ", stime = " << stime << ", vsize = " << vsize << ", rss = " << rss
-;
-  // return s << "utime = " << utime << ", vsize = " << vsize;
-  return os << "utime " << float(utime)/1.0e2 << "s, vsize " 
-	    << float(vsize)/1048576.0 << " Mb.";
+  // Don't fail if we can't read that (such as on a Mac), just return.
+  if (fp == NULL) {
+     return os;
+  } else {
+    int utime;
+    int stime;
+    unsigned int vsize;
+    unsigned int rss;
+    int result = 
+      fscanf(fp, "%*d %*s %*c %*d %*d %*d %*d %*d %*u %*u %*u %*u %*u %d %d %*d %*d %*d %*d"
+             "%*u %*u %*d %u %u", &utime, &stime, &vsize, &rss);
+    assert(result == 4);
+    fclose(fp);
+    // s << "utime = " << utime << ", stime = " << stime << ", vsize = " << vsize << ", rss = " << rss;
+    // return s << "utime = " << utime << ", vsize = " << vsize;
+    return os << "utime " << float(utime)/1.0e2 << "s, vsize " 
+	      << float(vsize)/1048576.0 << " Mb.";
+  }
 }
 #endif
 

second-stage/programs/features/utility.h

@@ -891,24 +891,42 @@ inline std::ostream& operator<< (std::ostream& os, const boost::shared_ptr<T>& s
 
 struct resource_usage { };
 
+#ifndef __i386
+#define NO_PROC_SELF_STAT
+#endif
+
+#ifdef __APPLE__
+#define NO_PROC_SELF_STAT
+#endif
+
+#ifdef NO_PROC_SELF_STAT
+inline std::ostream& operator<< (std::ostream& os, resource_usage r)
+{
+  return os;
+}
+#else // Assume we are on a 586 linux
 inline std::ostream& operator<< (std::ostream& os, resource_usage r)
 {
   FILE* fp = fopen("/proc/self/stat", "r");
-  assert(fp);
-  int utime;
-  int stime;
-  unsigned int vsize;
-  unsigned int rss;
-  int result = 
-    fscanf(fp, "%*d %*s %*c %*d %*d %*d %*d %*d %*u %*u %*u %*u %*u %d %d %*d %*d %*d %*d"
-           "%*u %*u %*d %u %u", &utime, &stime, &vsize, &rss);
-  assert(result == 4);
-  fclose(fp);
-  // s << "utime = " << utime << ", stime = " << stime << ", vsize = " << vsize << ", rss = " << rss
-;
-  // return s << "utime = " << utime << ", vsize = " << vsize;
-  return os << "utime " << float(utime)/1.0e2 << "s, vsize " 
-	    << float(vsize)/1048576.0 << " Mb.";
+  // Don't fail if we can't read that (such as on a Mac), just return.
+  if (fp == NULL) {
+     return os;
+  } else {
+    int utime;
+    int stime;
+    unsigned int vsize;
+    unsigned int rss;
+    int result = 
+      fscanf(fp, "%*d %*s %*c %*d %*d %*d %*d %*d %*u %*u %*u %*u %*u %d %d %*d %*d %*d %*d"
+             "%*u %*u %*d %u %u", &utime, &stime, &vsize, &rss);
+    assert(result == 4);
+    fclose(fp);
+    // s << "utime = " << utime << ", stime = " << stime << ", vsize = " << vsize << ", rss = " << rss;
+    // return s << "utime = " << utime << ", vsize = " << vsize;
+    return os << "utime " << float(utime)/1.0e2 << "s, vsize " 
+	      << float(vsize)/1048576.0 << " Mb.";
+  }
 }
+#endif
 
 #endif  // UTILITY_H