arffloader.java

矩阵的QR分解算法

    /**
     * Reads a single instance using the tokenizer and returns it. 
     *
     * @param structure 	the dataset header information, will get updated 
     * 				in case of string or relational attributes
     * @param flag 		if method should test for carriage return after 
     * 				each instance
     * @return 			null if end of file has been reached
     * @throws IOException 	if the information is not read 
     * successfully
     */ 
    public Instance readInstance(Instances structure, boolean flag) throws IOException {
      return getInstance(structure, flag);
    }
    
    /**
     * Reads a single instance using the tokenizer and returns it. 
     *
     * @param structure 	the dataset header information, will get updated 
     * 				in case of string or relational attributes
     * @param flag 		if method should test for carriage return after 
     * 				each instance
     * @return 			null if end of file has been reached
     * @throws IOException 	if the information is not read 
     * 				successfully
     */ 
    protected Instance getInstance(Instances structure, boolean flag) throws IOException {
      m_Data = structure;
      
      // Check if any attributes have been declared.
      if (m_Data.numAttributes() == 0) {
        errorMessage("no header information available");
      }

      // Check if end of file reached.
      getFirstToken();
      if (m_Tokenizer.ttype == StreamTokenizer.TT_EOF) {
        return null;
      }
      
      // Parse instance
      if (m_Tokenizer.ttype == '{') {
        return getInstanceSparse(flag);
      } else {
        return getInstanceFull(flag);
      }
    }

    /**
     * Reads a single instance using the tokenizer and returns it.
     *
     * @param flag 		if method should test for carriage return after 
     * 				each instance
     * @return 			null if end of file has been reached
     * @throws IOException 	if the information is not read 
     * 				successfully
     */ 
    protected Instance getInstanceSparse(boolean flag) throws IOException {
      int valIndex, numValues = 0, maxIndex = -1;
      
      // Get values
      do {
        // Get index
        getIndex();
        if (m_Tokenizer.ttype == '}') {
  	break;
        }
   
        // Is index valid?
        try{
  	m_IndicesBuffer[numValues] = Integer.valueOf(m_Tokenizer.sval).intValue();
        } catch (NumberFormatException e) {
  	errorMessage("index number expected");
        }
        if (m_IndicesBuffer[numValues] <= maxIndex) {
  	errorMessage("indices have to be ordered");
        }
        if ((m_IndicesBuffer[numValues] < 0) || 
  	  (m_IndicesBuffer[numValues] >= m_Data.numAttributes())) {
  	errorMessage("index out of bounds");
        }
        maxIndex = m_IndicesBuffer[numValues];

        // Get value;
        getNextToken();

        // Check if value is missing.
        if  (m_Tokenizer.ttype == '?') {
  	m_ValueBuffer[numValues] = Instance.missingValue();
        } else {

  	// Check if token is valid.
  	if (m_Tokenizer.ttype != StreamTokenizer.TT_WORD) {
  	  errorMessage("not a valid value");
  	}
          switch (m_Data.attribute(m_IndicesBuffer[numValues]).type()) {
            case Attribute.NOMINAL:
              // Check if value appears in header.
              valIndex = 
                m_Data.attribute(m_IndicesBuffer[numValues]).indexOfValue(m_Tokenizer.sval);
              if (valIndex == -1) {
                errorMessage("nominal value not declared in header");
              }
              m_ValueBuffer[numValues] = (double)valIndex;
              break;
  	case Attribute.NUMERIC:
  	  // Check if value is really a number.
  	  try{
  	    m_ValueBuffer[numValues] = Double.valueOf(m_Tokenizer.sval).
  	      doubleValue();
  	  } catch (NumberFormatException e) {
  	    errorMessage("number expected");
  	  }
            break;
  	case Attribute.STRING:
  	  m_ValueBuffer[numValues] = 
  	    m_Data.attribute(m_IndicesBuffer[numValues]).addStringValue(m_Tokenizer.sval);
            break;
          case Attribute.DATE:
            try {
              m_ValueBuffer[numValues] = 
                m_Data.attribute(m_IndicesBuffer[numValues]).parseDate(m_Tokenizer.sval);
            } catch (ParseException e) {
              errorMessage("unparseable date: " + m_Tokenizer.sval);
            }
            break;
          case Attribute.RELATIONAL:
            try {
              ArffReader arff = new ArffReader(new StringReader(m_Tokenizer.sval), m_Data.attribute(m_IndicesBuffer[numValues]).relation(), 0);
              Instances data = arff.getData();
              m_ValueBuffer[numValues] = m_Data.attribute(m_IndicesBuffer[numValues]).addRelation(data);
            }
            catch (Exception e) {
              throw new IOException(e.toString() + " of line " + getLineNo());
            }
            break;
          default:
            errorMessage("unknown attribute type in column " + m_IndicesBuffer[numValues]);
  	}
        }
        numValues++;
      } while (true);
      
      if (flag) {
        getLastToken(true);
      }
        
      // Add instance to dataset
      double[] tempValues = new double[numValues];
      int[] tempIndices = new int[numValues];
      System.arraycopy(m_ValueBuffer, 0, tempValues, 0, numValues);
      System.arraycopy(m_IndicesBuffer, 0, tempIndices, 0, numValues);
      Instance inst = new SparseInstance(1, tempValues, tempIndices, m_Data.numAttributes());
      inst.setDataset(m_Data);
      
      return inst;
    }

    /**
     * Reads a single instance using the tokenizer and returns it.
     *
     * @param flag 		if method should test for carriage return after 
     * 				each instance
     * @return 			null if end of file has been reached
     * @throws IOException 	if the information is not read 
     * 				successfully
     */ 
    protected Instance getInstanceFull(boolean flag) throws IOException {
      double[] instance = new double[m_Data.numAttributes()];
      int index;
      
      // Get values for all attributes.
      for (int i = 0; i < m_Data.numAttributes(); i++){
        // Get next token
        if (i > 0) {
  	getNextToken();
        }
              
        // Check if value is missing.
        if  (m_Tokenizer.ttype == '?') {
  	instance[i] = Instance.missingValue();
        } else {

  	// Check if token is valid.
  	if (m_Tokenizer.ttype != StreamTokenizer.TT_WORD) {
  	  errorMessage("not a valid value");
  	}
          switch (m_Data.attribute(i).type()) {
          case Attribute.NOMINAL:
  	  // Check if value appears in header.
  	  index = m_Data.attribute(i).indexOfValue(m_Tokenizer.sval);
  	  if (index == -1) {
  	    errorMessage("nominal value not declared in header");
  	  }
  	  instance[i] = (double)index;
            break;
  	case Attribute.NUMERIC:
  	  // Check if value is really a number.
  	  try{
  	    instance[i] = Double.valueOf(m_Tokenizer.sval).
  	      doubleValue();
  	  } catch (NumberFormatException e) {
  	    errorMessage("number expected");
  	  }
            break;
  	case Attribute.STRING:
  	  instance[i] = m_Data.attribute(i).addStringValue(m_Tokenizer.sval);
            break;
          case Attribute.DATE:
            try {
              instance[i] = m_Data.attribute(i).parseDate(m_Tokenizer.sval);
            } catch (ParseException e) {
              errorMessage("unparseable date: " + m_Tokenizer.sval);
            }
            break;
          case Attribute.RELATIONAL:
            try {
              ArffReader arff = new ArffReader(new StringReader(m_Tokenizer.sval), m_Data.attribute(i).relation(), 0);
              Instances data = arff.getData();
              instance[i] = m_Data.attribute(i).addRelation(data);
            }
            catch (Exception e) {
              throw new IOException(e.toString() + " of line " + getLineNo());
            }
            break;
          default:
            errorMessage("unknown attribute type in column " + i);
  	}
        }
      }
      
      if (flag) {
        getLastToken(true);
      }
        
      // Add instance to dataset
      Instance inst = new Instance(1, instance);
      inst.setDataset(m_Data);
      
      return inst;
    }

    /**
     * Reads and stores header of an ARFF file.
     *
     * @param capacity 		the number of instances to reserve in the data 
     * 				structure
     * @throws IOException 	if the information is not read 
     * 				successfully
     */ 
    protected void readHeader(int capacity) throws IOException {
      m_Lines = 0;
      String relationName = "";
      
      // Get name of relation.
      getFirstToken();
      if (m_Tokenizer.ttype == StreamTokenizer.TT_EOF) {
        errorMessage("premature end of file");
      }
      if (Instances.ARFF_RELATION.equalsIgnoreCase(m_Tokenizer.sval)) {
        getNextToken();
        relationName = m_Tokenizer.sval;
        getLastToken(false);
      } else {
        errorMessage("keyword " + Instances.ARFF_RELATION + " expected");
      }

      // Create vectors to hold information temporarily.
      FastVector attributes = new FastVector();
   
      // Get attribute declarations.
      getFirstToken();
      if (m_Tokenizer.ttype == StreamTokenizer.TT_EOF) {
        errorMessage("premature end of file");
      }

      while (Attribute.ARFF_ATTRIBUTE.equalsIgnoreCase(m_Tokenizer.sval)) {
        attributes = parseAttribute(attributes);
      }

      // Check if data part follows. We can't easily check for EOL.
      if (!Instances.ARFF_DATA.equalsIgnoreCase(m_Tokenizer.sval)) {
        errorMessage("keyword " + Instances.ARFF_DATA + " expected");
      }
      
      // Check if any attributes have been declared.
      if (attributes.size() == 0) {
        errorMessage("no attributes declared");
      }
      
      m_Data = new Instances(relationName, attributes, capacity);
    }

    /**
     * Parses the attribute declaration.
     *
     * @param attributes 		the current attributes vector
     * @return 			the new attributes vector
     * @throws IOException 	if the information is not read 
     * 				successfully
     */
    protected FastVector parseAttribute(FastVector attributes) throws IOException {
      String attributeName;
      FastVector attributeValues;

      // Get attribute name.
      getNextToken();
      attributeName = m_Tokenizer.sval;
      getNextToken();
      
      // Check if attribute is nominal.
      if (m_Tokenizer.ttype == StreamTokenizer.TT_WORD) {
        
        // Attribute is real, integer, or string.
        if (m_Tokenizer.sval.equalsIgnoreCase(Attribute.ARFF_ATTRIBUTE_REAL) ||
            m_Tokenizer.sval.equalsIgnoreCase(Attribute.ARFF_ATTRIBUTE_INTEGER) ||
            m_Tokenizer.sval.equalsIgnoreCase(Attribute.ARFF_ATTRIBUTE_NUMERIC)) {
          attributes.addElement(new Attribute(attributeName, attributes.size()));
          readTillEOL();
        } else if (m_Tokenizer.sval.equalsIgnoreCase(Attribute.ARFF_ATTRIBUTE_STRING)) {
          attributes.
            addElement(new Attribute(attributeName, (FastVector)null,
                attributes.size()));
          readTillEOL();
        } else if (m_Tokenizer.sval.equalsIgnoreCase(Attribute.ARFF_ATTRIBUTE_DATE)) {
          String format = null;
          if (m_Tokenizer.nextToken() != StreamTokenizer.TT_EOL) {
            if ((m_Tokenizer.ttype != StreamTokenizer.TT_WORD) &&
                (m_Tokenizer.ttype != '\'') &&
                (m_Tokenizer.ttype != '\"')) {
              errorMessage("not a valid date format");
            }
            format = m_Tokenizer.sval;
            readTillEOL();
          } else {
            m_Tokenizer.pushBack();
          }
          attributes.addElement(new Attribute(attributeName, format,
              attributes.size()));
          
        } else if (m_Tokenizer.sval.equalsIgnoreCase(Attribute.ARFF_ATTRIBUTE_RELATIONAL)) {
          readTillEOL();
          
          // Read attributes for subrelation
          // First, save current set of attributes