Class: SVMLight::Document

Inherits:

Object

• Object
• SVMLight::Document

Defined in:

lib/svmredlight/document.rb,
ext/svmredlight.c

Overview

A document is the Ruby representation of a DOC structure in SVMlight, it contains a queryid, a slackid, a costfactor ( c ) and a vector with feature numbers and their correspondent weights.

Class Method Summary

• .create(id, cost, slackid, queryid, words_ary) ⇒ Object

  Creates a DOC from an array of words it also takes an id -1 is normally OK for that value when using in filtering it also takes the C (cost) parameter for the SVM.

• .new(vector, opts = {}) ⇒ Object

Instance Method Summary

• #costfactor ⇒ Object
• #docnum ⇒ Object
• #queryid ⇒ Object
• #slackid ⇒ Object

Class Method Details

.create(id, cost, slackid, queryid, words_ary) ⇒ Object

Creates a DOC from an array of words it also takes an id -1 is normally OK for that value when using in filtering it also takes the C (cost) parameter for the SVM. words_ary an array of arrays like this

[wnum, weight], [wnum, weight], …

so we do not waste memory, defeating the svec implementation and do

not introduce a bunch of 0’s that seem to be OK when classifying but screw all up on training

# File 'ext/svmredlight.c', line 683

static VALUE
doc_create(VALUE klass, VALUE id, VALUE cost, VALUE slackid, VALUE queryid, VALUE words_ary ){
  long docnum, i, c_slackid, c_queryid;
  double c;
  WORD *words;
  SVECTOR *vec;
  DOC *d;
  VALUE inner_array;

  Check_Type(words_ary, T_ARRAY);
  Check_Type(slackid, T_FIXNUM);
  Check_Type(queryid, T_FIXNUM);

  if (RARRAY_LEN(words_ary) == 0)
    rb_raise(rb_eArgError, "Cannot create Document from empty arrays");

  words = (WORD*) my_malloc(sizeof(WORD) * (RARRAY_LEN(words_ary) + 1));

  for(i=0; i < (long)RARRAY_LEN(words_ary); i++){
    inner_array = RARRAY_PTR(words_ary)[i];
    Check_Type(inner_array, T_ARRAY);
    Check_Type(RARRAY_PTR(inner_array)[0], T_FIXNUM);

    if(!(TYPE(RARRAY_PTR(inner_array)[1]) == T_FLOAT ||  TYPE(RARRAY_PTR(inner_array)[1]) == T_FIXNUM ))
      rb_raise(rb_eArgError, "Feature weights must be numeric");
    
    if(FIX2LONG(RARRAY_PTR(inner_array)[0]) <= 0 )
      rb_raise(rb_eArgError, "Feature number has to be greater than zero");

    (words[i]).wnum     = FIX2LONG(RARRAY_PTR(inner_array)[0]);
    (words[i]).weight   = (FVAL)(NUM2DBL(RARRAY_PTR(inner_array)[1]));
  }
  words[i].wnum = 0;

  vec    = create_svector(words, (char*)"", 1.0);
  c      = NUM2DBL(cost);
  docnum = FIX2INT(id);

  d = create_example(docnum, FIX2LONG(queryid), FIX2LONG(slackid), c, vec);

  return Data_Wrap_Struct(klass, 0, doc_free, d);
}

.new(vector, opts = {}) ⇒ Object

Parameters:

• vector (Hash) —

  a hash where the keys are feature numbers and the values its weights

• opts (Hash) (defaults to: {}) —

  the options coincide with SVMLight parameters to the create_example function, the default values for all the options are 0

• [:docnum] (Hash) —

  a customizable set of options

• [:costfactor] (Hash) —

  a customizable set of options

• [:slackid] (Hash) —

  a customizable set of options

• [:queryid] (Hash) —

  a customizable set of options

# File 'lib/svmredlight/document.rb', line 11

def self.new(vector, opts={})
  opts.default = 0
  docnum     = opts[:docnum]
  costfactor = opts[:costfactor]
  slackid    = opts[:slackid] 
  queryid    = opts[:queryid] 

  create(docnum, costfactor, slackid, queryid, vector.to_a)
end

Instance Method Details

#costfactor ⇒ Object

# File 'ext/svmredlight.c', line 750

static VALUE
doc_get_costfactor(VALUE self){
  DOC *d;
  Data_Get_Struct(self, DOC, d);
 
  return DBL2NUM(d->costfactor);
}

#docnum ⇒ Object

# File 'ext/svmredlight.c', line 726

static VALUE
doc_get_docnum(VALUE self){
  DOC *d;
  Data_Get_Struct(self, DOC, d);
 
  return INT2FIX(d->docnum);
}

#queryid ⇒ Object

# File 'ext/svmredlight.c', line 742

static VALUE
doc_get_queryid(VALUE self){
  DOC *d;
  Data_Get_Struct(self, DOC, d);
 
  return INT2FIX(d->queryid);
}

#slackid ⇒ Object

# File 'ext/svmredlight.c', line 734

static VALUE
doc_get_slackid(VALUE self){
  DOC *d;
  Data_Get_Struct(self, DOC, d);
 
  return INT2FIX(d->slackid);
}