stanfordnlp/CoreNLP · error · RuntimeException

Unknown feature type " + feature

Error message

Unknown feature type " + feature

What it means

getFeatureTypeIndex maps a feature name to an index (0-5) based on its suffix (e.g. |C, |Cp2C, ... |Cp5C), which encodes the label-window size for that feature. If the feature name ends with none of the recognized suffixes, the classifier cannot determine how many labels the feature's weight matrix applies to, so it throws a RuntimeException.

Solutions

  1. Check the feature name reported in the exception; ensure both models were trained with the same featureFactory (flags.featureFactory) and its suffix conventions.
  2. Retrain or re-serialize both classifiers with matching feature definitions before calling combine().
  3. Extend getFeatureTypeIndex (or the feature factory) to emit a recognized suffix for your custom features, or exclude non-standard features before combining.

Example fix

// before: custom feature factory emits features like "MYFEAT-3"
// after: emit with the expected suffix so the type index resolves
new CoreLabel(); featName = base + "|C"; // e.g. "MYFEAT-3|C"
Defensive patterns

Strategy: validation

Validate before calling

List<String> SUFFIXES = List.of("|C", "|Cp2C", "|Cp3C", "|Cp4C", "|Cp5C");
for (String f : (Iterable<String>) crf.featureIndex) {
  if (SUFFIXES.stream().noneMatch(f::endsWith))
    throw new IllegalStateException("Unrecognized feature suffix: " + f);
}

Prevention

When it happens

Trigger: Calling combine() (or any code path that builds the feature-type map via getFeatureTypeIndex) when one classifier's featureIndex contains a feature whose name lacks a recognized '|CpNC' suffix; typically features created by a CRF with different featureFactory suffix configuration.

Common situations: Merging two CRF models trained with different feature factories or suffix conventions; custom SeqClassifierFlags.props / featureFactory strings that drop the standard suffix; loading a model whose feature naming doesn't match the current code's expectations.

Understand the failure class

Background: Invalid enum value errors: "Unknown type", "Invalid scope", "must be one of" — when a string is not on the library's allowed list — this error's family across 23 libraries.

Related errors

AI-assisted analysis of stanfordnlp/CoreNLP@1b7edd19c4 (2026-09-10). Data as JSON: /api/errors/4d874f598f585420. Report an issue: GitHub.

Appendix: source

Thrown at src/edu/stanford/nlp/ie/crf/CRFClassifier.java:257

   *
   * @param feature Feature string
   * @return index of featureType
   */
  private static int getFeatureTypeIndex(String feature) {
    if (feature.endsWith("|C")) {
      return 0;
    } else if (feature.endsWith("|CpC")) {
      return 1;
    } else if (feature.endsWith("|Cp2C")) {
      return 2;
    } else if (feature.endsWith("|Cp3C")) {
      return 3;
    } else if (feature.endsWith("|Cp4C")) {
      return 4;
    } else if (feature.endsWith("|Cp5C")) {
      return 5;
    } else {
      throw new RuntimeException("Unknown feature type " + feature);
    }
  }

  /**
   * Scales the weights of this CRFClassifier by the specified weight.
   *
   * @param scale The scale to multiply by
   */
  public void scaleWeights(double scale) {
    for (int i = 0; i < weights.length; i++) {
      for (int j = 0; j < weights[i].length; j++) {
        weights[i][j] *= scale;
      }
    }
  }

  /**
   * Combines weights from another crf (scaled by weight) into this CRF's

View on GitHub (pinned to 1b7edd19c4)