speech-tools/include/EST_Wagon.h
2015-09-19 10:52:26 +02:00

302 lines
12 KiB
C++

/*************************************************************************/
/* */
/* Centre for Speech Technology Research */
/* University of Edinburgh, UK */
/* Copyright (c) 1996,1997 */
/* All Rights Reserved. */
/* */
/* Permission is hereby granted, free of charge, to use and distribute */
/* this software and its documentation without restriction, including */
/* without limitation the rights to use, copy, modify, merge, publish, */
/* distribute, sublicense, and/or sell copies of this work, and to */
/* permit persons to whom this work is furnished to do so, subject to */
/* the following conditions: */
/* 1. The code must retain the above copyright notice, this list of */
/* conditions and the following disclaimer. */
/* 2. Any modifications must be clearly marked as such. */
/* 3. Original authors' names are not deleted. */
/* 4. The authors' names are not used to endorse or promote products */
/* derived from this software without specific prior written */
/* permission. */
/* */
/* THE UNIVERSITY OF EDINBURGH AND THE CONTRIBUTORS TO THIS WORK */
/* DISCLAIM ALL WARRANTIES WITH REGARD TO THIS SOFTWARE, INCLUDING */
/* ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS, IN NO EVENT */
/* SHALL THE UNIVERSITY OF EDINBURGH NOR THE CONTRIBUTORS BE LIABLE */
/* FOR ANY SPECIAL, INDIRECT OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES */
/* WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN */
/* AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, */
/* ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF */
/* THIS SOFTWARE. */
/* */
/*************************************************************************/
/* Author : Alan W Black */
/* Date : May 1996 */
/*-----------------------------------------------------------------------*/
/* */
/* Public declarations for Wagon (CART builder) */
/* */
/*=======================================================================*/
#ifndef __WAGON_H__
#define __WAGON_H__
#include "EST_String.h"
#include "EST_Val.h"
#include "EST_TVector.h"
#include "EST_TList.h"
#include "EST_simplestats.h" /* For EST_SuffStats class */
#include "EST_Track.h"
#include "siod.h"
#define wagon_error(WMESS) (cerr << WMESS << endl,exit(-1))
// I get floating point exceptions of Alphas when I do any comparisons
// with HUGE_VAL or FLT_MAX so I'll make my own
#define WGN_HUGE_VAL 1.0e20
class WVector : public EST_FVector
{
public:
WVector(int n) : EST_FVector(n) {}
int get_int_val(int n) const { return (int)a_no_check(n); }
float get_flt_val(int n) const { return a_no_check(n); }
void set_int_val(int n,int i) { a_check(n) = (int)i; }
void set_flt_val(int n,float f) { a_check(n) = f; }
};
typedef EST_TList<WVector *> WVectorList;
typedef EST_TVector<WVector *> WVectorVector;
/* Different types of feature */
enum wn_dtype {/* for predictees and predictors */
wndt_binary, wndt_float, wndt_class,
/* for predictees only */
wndt_cluster, wndt_vector, wndt_matrix, wndt_trajectory,
wndt_ols,
/* for ignored features */
wndt_ignore};
class WDataSet : public WVectorList {
private:
int dlength;
EST_IVector p_type;
EST_IVector p_ignore;
EST_StrVector p_name;
public:
void load_description(const EST_String& descfname,LISP ignores);
void ignore_non_numbers();
int ftype(const int &i) const {return p_type(i);}
int ignore(int i) const {return p_ignore(i); }
void set_ignore(int i,int value) { p_ignore[i] = value; }
const EST_String &feat_name(const int &i) const {return p_name(i);}
int samples(void) const {return length();}
int width(void) const {return dlength;}
};
enum wn_oper {wnop_equal, wnop_binary, wnop_greaterthan,
wnop_lessthan, wnop_is, wnop_in, wnop_matches};
class WQuestion {
private:
int feature_pos;
wn_oper op;
int yes;
int no;
EST_Val operand1;
EST_IList operandl;
float score;
public:
WQuestion() {;}
WQuestion(const WQuestion &s)
{ feature_pos=s.feature_pos;
op=s.op; yes=s.yes; no=s.no; operand1=s.operand1;
operandl = s.operandl; score=s.score;}
~WQuestion() {;}
WQuestion(int fp, wn_oper o,EST_Val a)
{ feature_pos=fp; op=o; operand1=a; }
void set_fp(const int &fp) {feature_pos=fp;}
void set_oper(const wn_oper &o) {op=o;}
void set_operand1(const EST_Val &a) {operand1 = a;}
void set_yes(const int &y) {yes=y;}
void set_no(const int &n) {no=n;}
int get_yes(void) const {return yes;}
int get_no(void) const {return no;}
const int get_fp(void) const {return feature_pos;}
const wn_oper get_op(void) const {return op;}
const EST_Val get_operand1(void) const {return operand1;}
const EST_IList &get_operandl(void) const {return operandl;}
const float get_score(void) const {return score;}
void set_score(const float &f) {score=f;}
const int ask(const WVector &w) const;
friend ostream& operator<<(ostream& s, const WQuestion &q);
};
enum wnim_type {wnim_unset, wnim_float, wnim_class,
wnim_cluster, wnim_vector, wnim_matrix, wnim_ols,
wnim_trajectory};
// Impurity measure for cumulating impurities from set of data
class WImpurity {
private:
wnim_type t;
EST_SuffStats a;
EST_DiscreteProbDistribution p;
float cluster_impurity();
float cluster_member_mean(int i);
float vector_impurity();
float trajectory_impurity();
float ols_impurity();
public:
EST_IList members; // Maybe there should be a cluster class
EST_FList member_counts; // AUP: Implement counts for vectors
EST_SuffStats **trajectory;
const WVectorVector *data; // Needed for ols
float score;
int l,width;
WImpurity() { t=wnim_unset; a.reset(); trajectory=0; l=0; width=0; data=0;}
~WImpurity();
WImpurity(const WVectorVector &ds);
void copy(const WImpurity &s)
{
int i,j;
t=s.t; a=s.a; p=s.p; members=s.members; member_counts = s.member_counts; l=s.l; width=s.width;
score = s.score;
data = s.data;
if (s.trajectory)
{
trajectory = new EST_SuffStats *[l];
for (i=0; i<l; i++)
{
trajectory[i] = new EST_SuffStats[width];
for (j=0; j<width; j++)
trajectory[i][j] = s.trajectory[i][j];
}
}
}
WImpurity &operator = (const WImpurity &a) { copy(a); return *this; }
float measure(void);
double samples(void);
wnim_type type(void) const { return t;}
void cumulate(const float pv,double count=1.0);
EST_Val value(void);
EST_DiscreteProbDistribution &pd() { return p; }
float cluster_distance(int i); // distance i from centre in sds
int in_cluster(int i); // distance i from centre < most remote member
float cluster_ranking(int i); // position in closeness to centre
friend ostream& operator<<(ostream &s, WImpurity &imp);
};
class WDlist {
private:
float p_score;
WQuestion p_question;
EST_String p_token;
int p_freq;
int p_samples;
WDlist *next;
public:
WDlist() { next=0; }
~WDlist() { if (next != 0) delete next; }
void set_score(float s) { p_score = s; }
void set_question(const WQuestion &q) { p_question = q; }
void set_best(const EST_String &t,int freq, int samples)
{ p_token = t; p_freq = freq; p_samples = samples;}
float score() const {return p_score;}
const EST_String &token(void) const {return p_token;}
const WQuestion &question() const {return p_question;}
EST_Val predict(const WVector &w);
friend WDlist *add_to_dlist(WDlist *l,WDlist *a);
friend ostream &operator<<(ostream &s, WDlist &d);
};
class WNode {
private:
WVectorVector data;
WQuestion question;
WImpurity impurity;
WNode *left;
WNode *right;
void print_out(ostream &s, int margin);
int leaf(void) const { return ((left == 0) || (right == 0)); }
int pure(void);
public:
WNode() { left = right = 0; }
~WNode() { if (left != 0) {delete left; left=0;}
if (right != 0) {delete right; right=0;} }
WVectorVector &get_data(void) { return data; }
void set_subnodes(WNode *l,WNode *r) { left=l; right=r; }
void set_impurity(const WImpurity &imp) {impurity=imp;}
void set_question(const WQuestion &q) {question=q;}
void prune(void);
void held_out_prune(void);
WImpurity &get_impurity(void) {return impurity;}
WQuestion &get_question(void) {return question;}
EST_Val predict(const WVector &w);
WNode *predict_node(const WVector &d);
int samples(void) const { return data.n(); }
friend ostream& operator<<(ostream &s, WNode &n);
};
extern Discretes wgn_discretes;
extern WDataSet wgn_dataset;
extern WDataSet wgn_test_dataset;
extern EST_FMatrix wgn_DistMatrix;
extern EST_Track wgn_VertexTrack;
extern EST_Track wgn_UnitTrack;
extern EST_Track wgn_VertexFeats;
void wgn_load_datadescription(EST_String fname,LISP ignores);
void wgn_load_dataset(WDataSet &ds,EST_String fname);
WNode *wgn_build_tree(float &score);
WNode *wgn_build_dlist(float &score,ostream *output);
WNode *wagon_stepwise(float limit);
float wgn_score_question(WQuestion &q, WVectorVector &ds);
void wgn_find_split(WQuestion &q,WVectorVector &ds,
WVectorVector &y,WVectorVector &n);
float summary_results(WNode &tree,ostream *output);
extern int wgn_min_cluster_size;
extern int wgn_held_out;
extern int wgn_prune;
extern int wgn_quiet;
extern int wgn_verbose;
extern int wgn_predictee;
extern int wgn_count_field;
extern EST_String wgn_count_field_name;
extern EST_String wgn_predictee_name;
extern float wgn_float_range_split;
extern float wgn_balance;
extern EST_String wgn_opt_param;
extern EST_String wgn_vertex_output;
#define wgn_ques_feature(X) (get_c_string(car(X)))
#define wgn_ques_oper_str(X) (get_c_string(car(cdr(X))))
#define wgn_ques_operand(X) (car(cdr(cdr(X))))
int wagon_ask_question(LISP question, LISP value);
int stepwise_ols(const EST_FMatrix &X,
const EST_FMatrix &Y,
const EST_StrList &feat_names,
float limit,
EST_FMatrix &coeffs,
const EST_FMatrix &Xtest,
const EST_FMatrix &Ytest,
EST_IVector &included,
float &best_score);
int robust_ols(const EST_FMatrix &X,
const EST_FMatrix &Y,
EST_IVector &included,
EST_FMatrix &coeffs);
int ols_apply(const EST_FMatrix &samples,
const EST_FMatrix &coeffs,
EST_FMatrix &res);
int ols_test(const EST_FMatrix &real,
const EST_FMatrix &predicted,
float &correlation,
float &rmse);
#endif /* __WAGON_H__ */