Solved some issues and prepared the library for using unordered_map instead of hash_map (C++ TR1). Can only use this when g++-4.0.3 is out

eo/contrib/mathsym/GNUmakefile

@@ -1,7 +1,7 @@
-COMPILEFLAGS=-Wno-deprecated -g -Wall -Wshadow #-DINTERVAL_DEBUG
-OPTFLAGS= -O3 -DNDEBUG 
+COMPILEFLAGS=-Wno-deprecated -g -Wall #-DINTERVAL_DEBUG
+OPTFLAGS= #-O3 -DNDEBUG 
 
-PROFILE_FLAGS=-pg
+PROFILE_FLAGS=#-pg 
 
 INCLUDES=-I. -Isym -Ifun -Igen -Ieval -Iregression -I../../src -Ieo_interface  -I..
 
@@ -33,7 +33,7 @@ distclean: clean
 	rm -rf tcc
 
 symreg: libsym.a symreg.o $(EXTLIBS)
-	$(CXX) -o symreg symreg.o libsym.a $(LIBS) $(PROFILE_FLAGS)
+	$(CXX) -o symreg symreg.o libsym.a $(LIBS) $(PROFILE_FLAGS) 
 
 libsym.a: $(OBJS) 
 	rm libsym.a; ar cq $(SYMLIB) $(OBJS) 

eo/contrib/mathsym/eo_interface/eoSym.h

@@ -33,20 +33,20 @@ class EoSym : public EO<Fitness>, public Sym {
 	    static_cast<Sym*>(this)->operator=(sym);
 	}
 	
-	virtual void printOn(ostream& os) const;
-	virtual void readFrom(istream& is);
+	virtual void printOn(std::ostream& os) const;
+	virtual void readFrom(std::istream& is);
 };
 
 
 template <class Fitness>
-void EoSym<Fitness>::printOn(ostream& os) const {
+void EoSym<Fitness>::printOn(std::ostream& os) const {
     EO<Fitness>::printOn(os);
     os << ' ';
     write_raw(os, *this);
 }
 
 template <class Fitness>
-void EoSym<Fitness>::readFrom(istream& is) {
+void EoSym<Fitness>::readFrom(std::istream& is) {
     EO<Fitness>::readFrom(is);
     read_raw(is, *this); 
 }
@@ -54,7 +54,7 @@ void EoSym<Fitness>::readFrom(istream& is) {
 template <class Fitness>
 inline std::ostream& operator<<(std::ostream& os, const EoSym<Fitness>& f) { f.printOn(os); return os; }
 template <class Fitness>
-inline istream& operator>>(std::istream& is, EoSym<Fitness>& f) { f.readFrom(is); return is; }
+inline std::istream& operator>>(std::istream& is, EoSym<Fitness>& f) { f.readFrom(is); return is; }
 
 
 #endif

eo/contrib/mathsym/eval/sym_compile.cpp

@@ -39,9 +39,14 @@ string make_prototypes() {
 }
 
 // contains variable names, like 'a0', 'a1', etc. or regular code
-typedef hash_map<Sym, string, HashSym> HashMap;
 
-HashMap::iterator find_entry(Sym sym, ostream& os, HashMap& map) {
+#if USE_TR1 
+typedef std::tr1::unordered_map<Sym, string, HashSym> HashMap;
+#else
+typedef hash_map<Sym, string, HashSym> HashMap;
+#endif
+
+HashMap::iterator find_entry(const Sym& sym, ostream& os, HashMap& map) {
     HashMap::iterator result = map.find(sym);
 
     if (result == map.end()) { // new entry
@@ -68,13 +73,82 @@ HashMap::iterator find_entry(Sym sym, ostream& os, HashMap& map) {
     return result;
 }
 
-void write_entry(Sym sym, ostream& os, HashMap& map, unsigned out) {
-    HashMap::iterator it = find_entry(sym, os, map);
+// prints 'num' in reverse notation. Does not matter as it's a unique id
+string make_var(unsigned num) {
+    string str = "a";
+    do {
+	str += char('0' + (num % 10));
+	num /= 10;
+    } while (num);
+    return str;
+}
+
+HashMap::iterator find_entry2(const Sym& sym, string& str, HashMap& map) {
+    HashMap::iterator result = map.find(sym);
+
+    if (result == map.end()) { // new entry
+	const SymVec& args = sym.args();
+	
+	vector<string> argstr(args.size());
+	for (unsigned i = 0; i < args.size(); ++i) {
+	    argstr[i] = find_entry2(args[i], str, map)->second;
+	}
+
+	string var = make_var(map.size()); // map.size(): unique id
+	
+	// write out the code
+	const FunDef& fun = get_element(sym.token());
+	string code = fun.c_print(argstr, vector<string>() );
+	str += "double " + var + "=" + code + ";\n";
+
+	result = map.insert( make_pair(sym, var ) ).first; // only want iterator
+    }
+    
+    return result;
+}
+
+void write_entry(const Sym& sym, ostream& os, HashMap& map, unsigned out) {
+    string s;
+    HashMap::iterator it = find_entry2(sym, s, map);
+    os << s;
     os << "y[" << out << "]=" << it->second << ";\n";
 }
 
+template <class T>
+string to_string(T t) {
+    ostringstream os;
+    os << t;
+    return os.str();
+}
+
+void write_entry(const Sym& sym, string& str, HashMap& map, unsigned out) {
+    HashMap::iterator it = find_entry2(sym, str, map);
+    
+    str += "y[" + to_string(out) + "]=" + it->second + ";\n";
+}
+
 multi_function compile(const std::vector<Sym>& syms) {
     
+    // static stream to avoid fragmentation of these LARGE strings
+    static string str;
+
+    str += make_prototypes();
+
+    str += "double func(const double* x, double* y) { \n ";
+   
+    HashMap map(Sym::get_dag().size());
+    
+    for (unsigned i = 0; i < syms.size(); ++i) {
+	write_entry(syms[i], str, map, i);
+    }
+    
+    str += ";}";
+ 
+    return (multi_function) symc_make(str.c_str(), "func"); 
+}
+
+multi_function compile2(const std::vector<Sym>& syms) {
+    
     // static stream to avoid fragmentation of these LARGE strings
     static ostringstream os;
     os.str("");
@@ -83,7 +157,8 @@ multi_function compile(const std::vector<Sym>& syms) {
 
     os << "double func(const double* x, double* y) { \n ";
    
-    HashMap map;
+    HashMap map(Sym::get_dag().size());
+    
     
     for (unsigned i = 0; i < syms.size(); ++i) {
 	write_entry(syms[i], os, map, i);
@@ -143,7 +218,7 @@ void compile(const std::vector<Sym>& syms, std::vector<single_function>& functio
     os.str("");
     
     os << make_prototypes();
-    HashMap map; 
+    HashMap map(Sym::get_dag().size());
     for (unsigned i = 0; i < syms.size(); ++i) {
 	
 	os << "double func" << i << "(const double* x) { return ";

eo/contrib/mathsym/fun/FunDef.cpp

@@ -120,7 +120,11 @@ struct HashDouble{
     }
 };
 
+#if USE_TR1
+typedef std::tr1::unordered_map<double, token_t> DoubleSet;
+#else
 typedef hash_map<double, token_t, HashDouble> DoubleSet;
+#endif
 
 static DoubleSet doubleSet; // for quick checking if a constant already exists
 static vector<bool> is_constant_flag;
@@ -186,6 +190,7 @@ bool is_constant(token_t token) {
 extern Sym default_const() { return SymConst(rng.normal()); }
 
 /* The functions */
+namespace {
 
 class Var : public FunDef {
     public :
@@ -247,7 +252,8 @@ class Const : public FunDef {
 
 	string name() const { return "parameter"; }
 };
-
+} // namespace 
+    
 void get_constants(Sym sym, vector<double>& ret) {
     token_t token = sym.token();
     if (is_constant_flag[token]) {
@@ -305,6 +311,12 @@ vector<const FunDef*> get_defined_functions() {
     return res;
 }
 
+FunDef* make_var(int idx) { return new Var(idx); }
+FunDef* make_const(double value) { return new Const(value); }
+
+
+namespace {
+
 class Sum : public FunDef {
 
     public :
@@ -478,6 +490,8 @@ struct Min {
     string name() const { return "min"; }
 };
 
+} // namespace
+
 string prototypes = "double pow(double, double);";
 string get_prototypes() { return prototypes; }
 unsigned add_prototype(string str) { prototypes += string("double ") + str + "(double);"; return prototypes.size(); }
@@ -488,6 +502,7 @@ token_t add_function(FunDef* function, token_t where) {
     return 0;
 }
 
+namespace {
 
 #define FUNCDEF(funcname) struct funcname##_struct { \
     double operator()(double val) const { return funcname(val); }\
@@ -498,9 +513,6 @@ token_t add_function(FunDef* function, token_t where) {
 static const token_t funcname##_token_static = add_function( new Unary<funcname##_struct>, funcname##_token);\
 unsigned funcname##_size = add_prototype(#funcname);
 
-FunDef* make_var(int idx) { return new Var(idx); }
-FunDef* make_const(double value) { return new Const(value); }
-
 static token_t ssum_token  = add_function( new Sum , sum_token);
 static token_t sprod_token = add_function( new Prod, prod_token);
 static token_t sinv_token  = add_function( new Unary<Inv>, inv_token);
@@ -524,11 +536,14 @@ FUNCDEF(atanh);
 
 FUNCDEF(exp);
 FUNCDEF(log);
+} // namespace
 
 double sqr(double x) { return x*x; }
 
+namespace {
 FUNCDEF(sqr);
 FUNCDEF(sqrt);
+} // namespace 
 
 /* Serialization */
 void write_raw(ostream& os, const Sym& sym) {

eo/contrib/mathsym/fun/FunDef.h

@@ -21,6 +21,7 @@
 #include <string>
 #include <vector>
 #include <memory>
+#include <iostream>
 
 #include "Sym.h"
 #include "Interval.h"
@@ -142,7 +143,7 @@ extern std::string get_prototypes();
 
 // reading and writing in internal format
 extern std::string write_raw(const Sym& sym);
-extern void write_raw(ostream& os, const Sym& sym);
+extern void write_raw(std::ostream& os, const Sym& sym);
 extern Sym read_raw(std::string str);
 extern Sym read_raw(std::istream& is); 
 extern void read_raw(std::istream& is, Sym& sym); 

eo/contrib/mathsym/regression/ErrorMeasure.cpp

@@ -232,8 +232,11 @@ class ErrorMeasureImpl {
 	vector<ErrorMeasure::result> calc_error(const vector<Sym>& pop) {
 
 	    // first declone
+#if USE_TR1
+	    typedef std::tr1::unordered_map<Sym, unsigned, HashSym> HashMap;
+#else
 	    typedef hash_map<Sym, unsigned, HashSym> HashMap;
-	    
+#endif	    
 	    HashMap clone_map;
 	    vector<Sym> decloned; 
 	    decloned.reserve(pop.size());

eo/contrib/mathsym/sym/Sym.cpp

@@ -46,7 +46,7 @@ size_t get_depth(const SymVec& vec) {
     return dp;
 }
 
-Sym::Sym(token_t tok, const SymVec& args_) 
+Sym::Sym(token_t tok, const SymVec& args_) : node(dag.end())
 {
     detail::SymKey key(tok, detail::SymArgs(args_));
     detail::SymValue val;
@@ -64,7 +64,7 @@ Sym::Sym(token_t tok, const SymVec& args_)
     else incref();
 }
 
-Sym::Sym(token_t tok, const Sym& a) { 
+Sym::Sym(token_t tok, const Sym& a) : node(dag.end()) { 
     SymVec args_; args_.push_back(a); 
     detail::SymKey key(tok, detail::SymArgs(args_));
     detail::SymValue val;
@@ -82,7 +82,7 @@ Sym::Sym(token_t tok, const Sym& a) {
     else incref();
 }
 
-Sym::Sym(token_t tok) {
+Sym::Sym(token_t tok) : node(dag.end()) {
     detail::SymKey key(tok);
     detail::SymValue val;
     node = dag.insert(pair<const detail::SymKey, detail::SymValue>(key, val)).first; 

eo/contrib/mathsym/sym/Sym.h

@@ -22,7 +22,7 @@
 
 #if __GNUC__ >= 3
 #include <backward/hash_map.h>
-#else
+#elif __GNUC__ < 3
 #include <hash_map.h>
 using std::hash_map;
 #endif
@@ -33,15 +33,23 @@ struct UniqueNodeStats { virtual ~UniqueNodeStats(){} };
 #include "SymImpl.h"
 #include "token.h"
 
-typedef hash_map<detail::SymKey, detail::SymValue, detail::SymKey::Hash> SymMap;
-typedef SymMap::iterator SymIterator;
-class Sym;
+#if __GNUC__ == 444
+#define USE_TR1 1
+#else
+#define USE_TR1 0
+#endif
 
-typedef std::vector<Sym> SymVec;
+#if USE_TR1
+#include <tr1/unordered_map>
+typedef std::tr1::unordered_map<detail::SymKey, detail::SymValue, detail::SymKey::Hash> SymMap;
+#else
+typedef hash_map<detail::SymKey, detail::SymValue, detail::SymKey::Hash> SymMap;
+#endif
+
+typedef SymMap::iterator SymIterator;
 
 /* Sym is the tree, for which all the nodes are stored in a hash table. 
  * This makes checking for equality O(1) */
-
 class Sym
 {
     public:
@@ -67,7 +75,7 @@ class Sym
 	/* Unique Stats are user defined */
 	UniqueNodeStats* extra_stats() const { return empty()? 0 : node->second.uniqueNodeStats; }
 	
-	int hashcode() const { detail::SymKey::Hash hash; return hash(node->first); }
+	int hashcode() const { return node->first.get_hash_code(); } //detail::SymKey::Hash hash; return hash(node->first); }
 	
 	// Friends, need to touch the node
 	friend struct detail::SymKey::Hash;

eo/contrib/mathsym/sym/SymImpl.cpp

@@ -21,7 +21,7 @@ namespace detail {
     
 class SymArgsImpl {
     public:
-    vector<Sym> owned_args;
+    std::vector<Sym> owned_args;
 };
 
 size_t SymArgs::len() const { 
@@ -76,7 +76,7 @@ int SymKey::calc_hash() const {
     unsigned long hash = unsigned(token);
     hash *= PRIMET;
     
-    const SymVec& v = args.vec();
+    const std::vector<Sym>& v = args.vec();
     for (unsigned i = 0; i < v.size(); ++i) {
 	hash += ( (v[i].address() >> 3) * primes[i%nprimes]) % HASHMOD;
     }

eo/contrib/mathsym/sym/SymImpl.h

@@ -22,6 +22,16 @@
 #include "token.h"
 
 class Sym;
+
+#if __GNUC__ > 4 || (__GNUC__ == 3 && __GNUC_MINOR__ >= 4)
+#include <ext/pool_allocator.h>
+typedef std::vector<Sym, __gnu_cxx::__pool_alloc<Sym> > std::vector<Sym>;
+//typedef std::vector<Sym> SymVec;
+#else
+typedef std::vector<Sym> SymVec;
+#endif
+
+
 namespace detail {
 
 class SymArgsImpl;
@@ -47,8 +57,8 @@ class SymArgs {
 class SymKey
 {
     public:
-	SymKey(token_t _token) : args(), token(_token) {}
-	SymKey(token_t _token, const detail::SymArgs& _args) : args(_args), token(_token) {}
+	SymKey(token_t _token) : args(), token(_token), hash_code(calc_hash()) {}
+	SymKey(token_t _token, const detail::SymArgs& _args) : args(_args), token(_token), hash_code(calc_hash()) {}
 	
 	
     private:
@@ -69,8 +79,11 @@ class SymKey
 	// fixates (i.e. claims memory) for the embedded vector of Syms
 	void fixate() const { args.fixate(); }
 	
+	int get_hash_code() const { return hash_code; }
+	
     private:
 	int calc_hash() const;
+	int hash_code;
 };
 
 struct SymValue

eo/contrib/mathsym/symreg.cpp

@@ -306,6 +306,7 @@ int main(int argc, char* argv[]) {
     genmon.add(avgSize);
     genmon.add(dagSize);
     genmon.add(sumSize);
+    genmon.add(term); // add generation counter
     
     checkpoint.add(genmon);