userclassifier.java

:<<数据挖掘--实用机器学习技术及java实现>>一书的配套源程序

     * This function gets called to set the node to use a linear regression
     * and attribute filter.
     * @exception If can't set a default linear egression model.
     */
    private void setLinear() throws Exception {
      //then set default behaviour for node.
      //set linear regression combined with attribute filter
      
      //find the attributes used for splitting.
      boolean[] attributeList = new boolean[m_training.numAttributes()];
      for (int noa = 0; noa < m_training.numAttributes(); noa++) {
	attributeList[noa] = false;
      }
      
      TreeClass temp = this;
      attributeList[m_training.classIndex()] = true;
      while (temp != null) {
	attributeList[temp.m_attrib1] = true;
	attributeList[temp.m_attrib2] = true;
	temp = temp.m_parent;
      }
      int classind = 0;
      
      
      //find the new class index
      for (int noa = 0; noa < m_training.classIndex(); noa++) {
	if (attributeList[noa]) {
	  classind++;
	}
      }
      //count how many attribs were used
      int count = 0;
      for (int noa = 0; noa < m_training.numAttributes(); noa++) {
	if (attributeList[noa]) {
	  count++;
	}
      }
      
      
      //fill an int array with the numbers of those attribs
      int[] attributeList2 = new int[count];
      count = 0;
      for (int noa = 0; noa < m_training.numAttributes(); noa++) {
	if (attributeList[noa]) {
	  attributeList2[count] = noa;
	  count++;
	}
      }
      
      m_filter = new AttributeFilter();
      ((AttributeFilter)m_filter).setInvertSelection(true);
      ((AttributeFilter)m_filter).setAttributeIndicesArray(attributeList2);
      m_filter.setInputFormat(m_training);
      
      Instances temp2 = Filter.useFilter(m_training, m_filter);
      temp2.setClassIndex(classind);
      m_classObject = new LinearRegression();
      m_classObject.buildClassifier(temp2);
      
      
      
    }
    
    
    /**
     * Call to find out if an instance is in a polyline.
     * @param ob The polyline to check.
     * @param x The value of attribute1 to check.
     * @param y The value of attribute2 to check.
     * @return True if inside, false if not.
     */
    private boolean inPolyline(FastVector ob, double x, double y) {
      //this works similar to the inPoly below except that
      //the first and last lines are treated as extending infinite 
      //in one direction and 
      //then infinitly in the x dirction their is a line that will 
      //normaly be infinite but
      //can be finite in one or both directions
      
      int countx = 0;
      double vecx, vecy;
      double change;
      double x1, y1, x2, y2;
      
      for (int noa = 1; noa < ob.size() - 4; noa+= 2) {
	y1 = ((Double)ob.elementAt(noa+1)).doubleValue();
	y2 = ((Double)ob.elementAt(noa+3)).doubleValue();
	x1 = ((Double)ob.elementAt(noa)).doubleValue();
	x2 = ((Double)ob.elementAt(noa+2)).doubleValue();
	vecy = y2 - y1;
	vecx = x2 - x1;
	if (noa == 1 && noa == ob.size() - 6) {
	  //then do special test first and last edge
	  if (vecy != 0) {
	    change = (y - y1) / vecy;
	    if (vecx * change + x1 >= x) {
	      //then intersection
	      countx++;
	    }
	  }
	  
	  
	}
	else if (noa == 1) {
	  if ((y < y2 && vecy > 0) || (y > y2 && vecy < 0)) {
	    //now just determine intersection or not
	    change = (y - y1) / vecy;
	    if (vecx * change + x1 >= x) {
	      //then intersection on horiz
	      countx++;
	    }
	  }
	}
	else if (noa == ob.size() - 6) {
	  //then do special test on last edge
	  if ((y <= y1 && vecy < 0) || (y >= y1 && vecy > 0)) {
	    change = (y - y1) / vecy;
	    if (vecx * change + x1 >= x) {
	      countx++;
	    }
	  }
	  
	}
	else if ((y1 <= y && y < y2) || (y2 < y && y <= y1)) {
	  //then continue tests.
	  if (vecy == 0) {
	    //then lines are parallel stop tests in 
	    //ofcourse it should never make it this far
	  }
	  else {
	    change = (y - y1) / vecy;
	    if (vecx * change + x1 >= x) {
	      //then intersects on horiz
	      countx++;
	    }
	  }
	}
	
      }
      
      //now check for intersection with the infinity line
      y1 = ((Double)ob.elementAt(ob.size() - 2)).doubleValue();
      y2 = ((Double)ob.elementAt(ob.size() - 1)).doubleValue();
      
      if (y1 > y2) {
	//then normal line
	if (y1 >= y && y > y2) {
	  countx++;
	}
      }
      else {
	//then the line segment is inverted
	if (y1 >= y || y > y2) {
	  countx++;
	}
      }
      
      
      if ((countx % 2) == 1) {
	return true;
      }
      else {
	return false;
      }
      
      
    }
    
    
    /** 
     * Call this to determine if an instance is in a polygon.
     * @param ob The polygon.
     * @param x The value of attribute 1.
     * @param y The value of attribute 2.
     * @return True if in polygon, false if not.
     */
    private boolean inPoly(FastVector ob, double x, double y) {
      int count = 0;
      double vecx, vecy;
      double change;
      double x1, y1, x2, y2;
      for (int noa = 1; noa < ob.size() - 2; noa += 2) {
	y1 = ((Double)ob.elementAt(noa+1)).doubleValue();
	y2 = ((Double)ob.elementAt(noa+3)).doubleValue();
	if ((y1 <= y && y < y2) || (y2 < y && y <= y1)) {
	  //then continue tests.
	  vecy = y2 - y1;
	  if (vecy == 0) {
	    //then lines are parallel stop tests for this line
	  }
	  else {
	    x1 = ((Double)ob.elementAt(noa)).doubleValue();
	    x2 = ((Double)ob.elementAt(noa+2)).doubleValue();
	    vecx = x2 - x1;
	    change = (y - y1) / vecy;
	    if (vecx * change + x1 >= x) {
	      //then add to count as an intersected line
	      count++;
	    }
	  }
	  
	}
      }
      if ((count % 2) == 1) {
	//then lies inside polygon
	//System.out.println("in");
	return true;
      }
      else {
	//System.out.println("out");
	return false;
      }
      //System.out.println("WHAT?!?!?!?!!?!??!?!");
      //return false;
    }
    

    /**
     * Goes through the tree structure recursively and returns the node that
     * has the id.
     * @param id The node to find.
     * @return The node that matches the id.
     */
    public TreeClass getNode(String id) {
      //returns the treeclass object with the particular ident
      if (id.equals(m_identity)) {
	return this;
      }
      
      if (m_set1 != null) {
	TreeClass tmp = m_set1.getNode(id);
	if (tmp != null) {
	  return tmp;
	}
      }
      if (m_set2 != null) {
	TreeClass tmp = m_set2.getNode(id);
	if (tmp != null) {
	  return tmp;
	}
      }
      return null;
    }
    
    
    /**
     * Returns a string containing a bit of information about this node, in 
     * alternate form.
     * @param s The string buffer to fill.
     * @exception Exception if can't create label.
     */
    public void getAlternateLabel(StringBuffer s) throws Exception {
      
      //StringBuffer s = new StringBuffer();
      
      FastVector tmp = (FastVector)m_ranges.elementAt(0);
      
      if (m_classObject != null && m_training.classAttribute().isNominal()) {
	s.append("Classified by " + m_classObject.getClass().getName());
      }
      else if (((Double)tmp.elementAt(0)).intValue() == LEAF) {
	if (m_training.classAttribute().isNominal()) {
	  double high = -1000;
	  int num = 0;
	  double count = 0;
	  for (int noa = 0; noa < m_training.classAttribute().numValues();
	       noa++) {
	    if (((Double)tmp.elementAt(noa + 1)).doubleValue() > high) {
	      high = ((Double)tmp.elementAt(noa + 1)).doubleValue();
	      num  = noa + 1;
	    }
	    count += ((Double)tmp.elementAt(noa + 1)).doubleValue();
	  }
	  s.append(m_training.classAttribute().value(num-1) + "(" + count);
	  if (count > high) {
	    s.append("/" + (count - high));
	  }
	  s.append(")");
	}
	else {
	  if (m_classObject == null 
	      && ((Double)tmp.elementAt(0)).intValue() == LEAF) {
	    setLinear();
	  }
	  s.append("Standard Deviation = " 
		   + Utils.doubleToString(((Double)tmp.elementAt(1))
					  .doubleValue(), 6));
	  
	}
      }
      else {
	s.append("Split on ");
	s.append(m_training.attribute(m_attrib1).name() + " AND ");
	s.append(m_training.attribute(m_attrib2).name());
	
	
      }
      
      //return s.toString();
    }
    

    
    /**
     * Returns a string containing a bit of information about this node.
     * @param s The stringbuffer to fill.
     * @exception Exception if can't create label.
     */
    public void getLabel(StringBuffer s) throws Exception {
      //for now just return identity
      //StringBuffer s = new StringBuffer();
      
      FastVector tmp = (FastVector)m_ranges.elementAt(0);
      
      
      if (m_classObject != null && m_training.classAttribute().isNominal()) {
	s.append("Classified by\\n" + m_classObject.getClass().getName());
      }
      else if (((Double)tmp.elementAt(0)).intValue() == LEAF) {
	
	if (m_training.classAttribute().isNominal()) {
	  boolean first = true;
	  for (int noa = 0; noa < m_training.classAttribute().numValues(); 
	       noa++) {
	    if (((Double)tmp.elementAt(noa + 1)).doubleValue() > 0) {
	      if (first)
		{
		  s.append("[" + m_training.classAttribute().value(noa));
		  first = false;
		}
	      else
		{
		  s.append("\\n[" + m_training.classAttribute().value(noa));
		}
	      s.append(", " + ((Double)tmp.elementAt(noa + 1)).doubleValue() 
		       + "]");
	    }      
	  }
	}
	else {
	  if (m_classObject == null 
	      && ((Double)tmp.elementAt(0)).intValue() == LEAF) {
	    setLinear();
	  }
	  s.append("Standard Deviation = " 
		   + Utils.doubleToString(((Double)tmp.elementAt(1))
		   .doubleValue(), 6));
	}
      }
      else {
	s.append("Split on\\n");
	s.append(m_training.attribute(m_attrib1).name() + " AND\\n");
	s.append(m_training.attribute(m_attrib2).name());
      }
      //return s.toString();
    }

    /**
     * Converts The tree structure to a dotty string.
     * @param t The stringbuffer to fill with the dotty structure.
     * @exception Exception if can't convert structure to dotty.
     */
    public void toDotty(StringBuffer t) throws Exception {
      //this will recursively create all the dotty info for the structure
      t.append(m_identity + " [label=\"");
      getLabel(t);
      t.append("\" ");
      //System.out.println(((Double)((FastVector)ranges.elementAt(0)).
      //elementAt(0)).intValue() + " A num ");
      if (((Double)((FastVector)m_ranges.elementAt(0)).elementAt(0)).intValue()
	  == LEAF) {
	t.append("shape=box ");
      }
      else {
	t.append("shape=ellipse ");
      }
      t.append("style=filled color=gray95]\n");
      
      if (m_set1 != null) {
	t.append(m_identity + "->");
	t.append(m_set1.m_identity + " [label=\"True\"]\n");//the edge for 
	//the left
	m_set1.toDotty(t);
      }
      if (m_set2 != null) {
	t.append(m_identity + "->");
	t.append(m_set2.m_identity + " [label=\"False\"]\n"); //the edge for 
	//the 
	//right
	m_set2.toDotty(t);
      }
      
    }
    
    /**
     * This will append the class Object in the tree to the string buffer.
     * @param t The stringbuffer.
     */
    public void objectStrings(StringBuffer t) {
      
      if (m_classObject != null) {
	t.append("\n\n" + m_identity +" {\n" + m_classObject.toString()+"\n}");
      }
      if (m_set1 != null) {
	m_set1.objectStrings(t);
      }
      if (m_set2 != null) {
	m_set2.objectStrings(t);
      }
    }
    
    /**
     * Converts the tree structure to a string. for people to read.
     * @param l How deep this node is in the tree.
     * @param t The stringbuffer to fill with the string.
     * @exception Exception if can't convert th string.
     */
    public void toString(int l, StringBuffer t) throws Exception {
      

      
      if (((Double)((FastVector)m_ranges.elementAt(0)).elementAt(0)).intValue()
	  == LEAF) {
	t.append(": " + m_identity + " ");
	getAlternateLabel(t);
      }
      if (m_set1 != null) {
	t.append("\n");
	for (int noa = 0; noa < l; noa++) {
	  t.append("|   ");
	  
	}
	getAlternateLabel(t);
	t.append(" (In Set)");
	m_set1.toString(l+1, t);
      }
      if (m_set2 != null) {
	t.append("\n");
	for (int noa = 0; noa < l; noa++) {
	  t.append("|   ");
	}
	getAlternateLabel(t);
	t.append(" (Not in Set)");
	m_set2.toString(l+1, t);
      }
      //return t.toString();
    }
    
  }
  
  
  
}