Package com.sun.speech.freetts.cart

Class CARTImpl

java.lang.Object
com.sun.speech.freetts.cart.CARTImpl
All Implemented Interfaces:
CART
public class CARTImpl extends Object implements CART
Implementation of a Classification and Regression Tree (CART) that is used more like a binary decision tree, with each node containing a decision or a final value. The decision nodes in the CART trees operate on an Item and have the following format: 
   NODE feat operand value qfalse

Where feat is an string that represents a feature to pass to the findFeature method of an item.

The value represents the value to be compared against the feature obtained from the item via the feat string. The operand is the operation to do the comparison. The available operands are as follows:

  • < - the feature is less than value
  • = - the feature is equal to the value
  • > - the feature is greater than the value
  • MATCHES - the feature matches the regular expression stored in value
  • IN - [[[TODO: still guessing because none of the CART's in Flite seem to use IN]]] the value is in the list defined by the feature.

[[[TODO: provide support for the IN operator.]]]

For < and >, this CART coerces the value and feature to float's. For =, this CART coerces the value and feature to string and checks for string equality. For MATCHES, this CART uses the value as a regular expression and compares the obtained feature to that.

A CART is represented by an array in this implementation. The qfalse value represents the index of the array to go to if the comparison does not match. In this implementation, qtrue index is always implied, and represents the next element in the array. The root node of the CART is the first element in the array.

The interpretations always start at the root node of the CART and continue until a final node is found. The final nodes have the following form: 

   LEAF value

Where value represents the value of the node. Reaching a final node indicates the interpretation is over and the value of the node is the interpretation result.

  • Constructor Details

    • CARTImpl

      public CARTImpl(URL url) throws IOException
      Creates a new CART by reading from the given URL.
      Parameters:
      url - the location of the CART data
      Throws:
      IOException - if errors occur while reading the data

    • CARTImpl

      public CARTImpl(BufferedReader reader, int nodes) throws IOException
      Creates a new CART by reading from the given reader.
      Parameters:
      reader - the source of the CART data
      nodes - the number of nodes to read for this cart
      Throws:
      IOException - if errors occur while reading the data

  • Method Details

    • dumpBinary

      public void dumpBinary(DataOutputStream os) throws IOException
      Dumps this CART to the output stream.
      Specified by:
      dumpBinary in interface CART
      Parameters:
      os - the output stream
      Throws:
      IOException - if an error occurs during output

    • loadBinary

      public static CART loadBinary(ByteBuffer bb) throws IOException
      Loads a CART from the input byte buffer.
      Parameters:
      bb - the byte buffer
      Returns:
      the CART
      Throws:
      IOException - if an error occurs during output Note that cart nodes are really saved as strings that have to be parsed.

    • loadBinary

      public static CART loadBinary(DataInputStream is) throws IOException
      Loads a CART from the input stream.
      Parameters:
      is - the input stream
      Returns:
      the CART
      Throws:
      IOException - if an error occurs during output Note that cart nodes are really saved as strings that have to be parsed.

    • parseAndAdd

      protected void parseAndAdd(String line)
      Creates a node from the given input line and add it to the CART. It expects the TOTAL line to come before any of the nodes.
      Parameters:
      line - a line of input to parse

    • getNode

      protected com.sun.speech.freetts.cart.CARTImpl.Node getNode(String type, StringTokenizer tokenizer, int currentNode)
      Gets the node based upon the type and tokenizer.
      Parameters:
      type - NODE or LEAF
      tokenizer - the StringTokenizer containing the data to get
      currentNode - the index of the current node we're looking at
      Returns:
      the node

    • parseValue

      protected Object parseValue(String string)
      Coerces a string into a value.
      Parameters:
      string - of the form "type(value)"; for example, "Float(2.3)"
      Returns:
      the value

    • interpret

      public Object interpret(Item item)
      Passes the given item through this CART and returns the interpretation.
      Specified by:
      interpret in interface CART
      Parameters:
      item - the item to analyze
      Returns:
      the interpretation