Examples of edu.stanford.nlp.sequences.SeqClassifierFlags

edu.stanford.nlp.sequences.SeqClassifierFlags

Flags for sequence classifiers. Documentation for general flags and flags for NER can be found in the Javadoc of {@link edu.stanford.nlp.ie.NERFeatureFactory}. Documentation for the flags for Chinese word segmentation can be found in the Javadoc of {@link edu.stanford.nlp.wordseg.ChineseSegmenterFeatureFactory}.
IMPORTANT NOTE IF CHANGING THIS FILE: MAKE SURE TO ONLY ADD NEW VARIABLES AT THE END OF THE LIST OF VARIABLES (and not to change existing variables)! Otherwise you usually break all currently serialized classifiers!!! Search for "ADD VARIABLES ABOVE HERE" below.
Some general flags are described here StringString

Property Name	Type	Default Value	Description
useQN	boolean	true	Use Quasi-Newton (L-BFGS) optimization to find minimum. NOTE: Need to set this to false if using other minimizers such as SGD.
QNsize	int	25	Number of previous iterations of Quasi-Newton to store (this increases memory use, but speeds convergence by letting the Quasi-Newton optimization more effectively approximate the second derivative).
QNsize2	int	25	Number of previous iterations of Quasi-Newton to store (used when pruning features, after the first iteration - the first iteration is with QNSize).
useInPlaceSGD	boolean	false	Use SGD (tweaking weights in place) to find minimum (more efficient than the old SGD, faster to converge than Quasi-Newton if there are very large of samples). Implemented for CRFClassifier. NOTE: Remember to set useQN to false
tuneSampleSize	int	-1	If this number is greater than 0, specifies the number of samples to use for tuning (default is 1000).
SGDPasses	int	-1	If this number is greater than 0, specifies the number of SGD passes over entire training set) to do before giving up (default is 50). Can be smaller if sample size is very large.
useSGD	boolean	false	Use SGD to find minimum (can be slow). NOTE: Remember to set useQN to false
useSGDtoQN	boolean	false	Use SGD (SGD version selected by useInPlaceSGD or useSGD) for a certain number of passes (SGDPasses) and then switches to QN. Gives the quick initial convergence of SGD, with the desired convergence criterion of QN (there is some rampup time for QN). NOTE: Remember to set useQN to false
evaluateIters	int	0	If this number is greater than 0, evaluates on the test set every so often while minimizing. Implemented for CRFClassifier.
evalCmd	String		If specified (and evaluateIters is set), runs the specified cmdline command during evaluation (instead of default CONLL-like NER evaluation)
evaluateTrain	boolean	false	If specified (and evaluateIters is set), also evaluate on training set (can be expensive)
tokenizerOptions	(null)	Extra options to supply to the tokenizer when creating it.
tokenizerFactory	(null)	A different tokenizer factory to use if the ReaderAndWriter in question uses tokenizers.

@author Jenny Finkel


  /** Default place to look in Jar file for classifier. */
  public static final String DEFAULT_CLASSIFIER = "/classifiers/ner-eng-ie.cmm-3-all2006.ser.gz";


  protected CMMClassifier() {
    super(new SeqClassifierFlags());
  }

View Full Code Here

  }


  public static void main(String[] args) {
    Properties props = StringUtils.argsToProperties(args);
    // System.err.println(props.toString());
    SeqClassifierFlags flags = new SeqClassifierFlags(props);
    MaxMatchSegmenter seg = new MaxMatchSegmenter();
    String lexiconFile = props.getProperty("lexicon");
    if(lexiconFile != null) {
      seg.addLexicon(lexiconFile);
    } else {

View Full Code Here

    props.remove(optWithDomains);
    props.remove(optDomain);
    props.remove(optNoRewrites);
    props.remove(optLocalFeaturesOnly);


    flags = new SeqClassifierFlags(props);
    classifier = new CRFClassifier<CoreLabel>(flags);
  }

View Full Code Here


    Properties props = new Properties();
    props.setProperty("useGazettes", "true");
    props.setProperty("sloppyGazette", "true");
    props.setProperty("gazette", "projects/core/data/edu/stanford/nlp/ie/test_gazette.txt");
    SeqClassifierFlags flags = new SeqClassifierFlags(props);
    NERFeatureFactory<CoreLabel> factory = new NERFeatureFactory<CoreLabel>();
    factory.init(flags);


    Set<String> features;
    features = new HashSet<String>(factory.featuresC(paddedSentence, 4));

View Full Code Here

TOP

Related Classes of edu.stanford.nlp.sequences.SeqClassifierFlags

edu.stanford.nlp.ie.ner.CMMClassifier

edu.stanford.nlp.ie.NERFeatureFactoryITest

edu.stanford.nlp.international.arabic.process.ArabicSegmenter

edu.stanford.nlp.wordseg.MaxMatchSegmenter

All source code are property of their respective owners. Java is a trademark of Sun Microsystems, Inc and owned by ORACLE Inc. Contact coftware#gmail.com.