stanfordnlp/CoreNLP · error · RuntimeException

linearConstraints.length (

Error message

linearConstraints.length (

What it means

ExactBestSequenceFinder.bestSequence performs Viterbi decoding over a SequenceModel. If linearConstraints is supplied, it must have one entry per padded position (length + leftWindow + rightWindow). A mismatch throws RuntimeException describing the lengths.

Solutions

1. Size the constraints array as ts.length() + ts.leftWindow() + ts.rightWindow() before calling bestSequence.
2. Pad an existing constraints array with default (unconstrained, e.g. null/empty) entries for the window positions.
3. Pass null if no constraints are needed instead of a wrongly sized array.

Example fix

// before
int[] constraints = new int[ts.length()];
new ExactBestSequenceFinder().bestSequence(ts, constraints);
// after
int padLength = ts.length() + ts.leftWindow() + ts.rightWindow();
int[] constraints = new int[padLength];
new ExactBestSequenceFinder().bestSequence(ts, constraints);

// size constraints to the padded length before decoding
int padLength = ts.length() + ts.leftWindow() + ts.rightWindow();
if (linearConstraints != null && linearConstraints.length != padLength)
  linearConstraints = java.util.Arrays.copyOf(linearConstraints, padLength);

When it happens

Trigger: Calling bestSequence(ts, linearConstraints) where linearConstraints.length != ts.length() + ts.leftWindow() + ts.rightWindow() — e.g. constraints sized only by the raw sequence length while the model has non-zero windows, at ExactBestSequenceFinder.java:48.

Common situations: Building constraint arrays for the un-padded document length in NER/tagger pipelines while the test sequence includes left/right context windows; reusing constraints across models with different window sizes.

Understand the failure class

Background: Tensor shape mismatch errors ("must have shape", "expected shape ... got ..."): when tensor dimensions disagree with what an op or layer was told to expect — this error's family across 6 libraries.

Related errors

• input must be sorted!
• Two models must have the same number of classes
• Two models must have the same sequence length
• KBestSequenceFinder only works with rightWindow == 0 not
• Unknown feature type " + feature

AI-assisted analysis of stanfordnlp/CoreNLP@1b7edd19c4 (2026-09-10). Data as JSON: /api/errors/b28aaf0600690602. Report an issue: GitHub.

Appendix: source

Thrown at src/edu/stanford/nlp/sequences/ExactBestSequenceFinder.java:48

   * Runs the Viterbi algorithm on the sequence model given by the TagScorer
   * in order to find the best sequence.
   *
   * @param ts The SequenceModel to be used for scoring
   * @return An array containing the int tags of the best sequence
   */
  @Override
  public int[] bestSequence(SequenceModel ts) {
    return bestSequence(ts, null).first();
  }

  private static Pair<int[], Double> bestSequence(SequenceModel ts, double[][] linearConstraints) {
    // Set up tag options
    final int length = ts.length();
    final int leftWindow = ts.leftWindow();
    final int rightWindow = ts.rightWindow();
    final int padLength = length + leftWindow + rightWindow;
    if (linearConstraints != null && linearConstraints.length != padLength)
      throw new RuntimeException("linearConstraints.length (" +  linearConstraints.length + ") does not match padLength (" + padLength + ") of SequenceModel" + ", length=="+length+", leftW="+leftWindow+", rightW="+rightWindow);
    int[][] tags = new int[padLength][];
    int[] tagNum = new int[padLength];
    if (DEBUG) { log.info("Doing bestSequence length " + length + "; leftWin " + leftWindow + "; rightWin " + rightWindow + "; padLength " + padLength); }
    for (int pos = 0; pos < padLength; pos++) {
      // potentially constrain values considered in inference (e.g., to only observed tags for a word if word is common)
      tags[pos] = ts.getPossibleValues(pos);
      tagNum[pos] = tags[pos].length;
      if (DEBUG) { log.info("There are " + tagNum[pos] + " values at position " + pos + ": " + Arrays.toString(tags[pos])); }
    }

    // Set up product space sizes
    int[] productSizes = initProductSizes(ts, tagNum, new int[padLength]);

    // Score all of each window's options
    int[] tempTags = new int[padLength];
    double[][] windowScore = computeWindowScore(ts, tags, tagNum, tempTags, productSizes);

    // Set up score and backtrace arrays

View on GitHub (pinned to 1b7edd19c4)