stanfordnlp/CoreNLP · error · IllegalStateException

neg log lik smaller than 0: " + s

Error message

neg log lik smaller than 0: " + s

What it means

LambdaSolve's log-likelihood computation (2LogLikelihood / related CG variant) asserts an internal invariant: the negative log likelihood over the model must never be negative. A negative value means the computed conditional probabilities are inconsistent, so it throws IllegalStateException.

Solutions

1. Verify the lambdas were produced for the SAME problem instance (same features and data) being evaluated
2. Re-train the model instead of loading questionable lambdas
3. Sanity-check lambda values (magnitude, NaN) before computing likelihood

Example fix

// before
double ll = problem.twoLogLikelihood(lambdasFromOtherProblem);
// after
if (lambdasFromOtherProblem.length != problem.fValues.length) {
  throw new IllegalArgumentException("Lambda vector does not match this problem");
}
double ll = problem.twoLogLikelihood(lambdasFromOtherProblem);

if (lambdas.length != expectedLambdaCount || Arrays.stream(lambdas).anyMatch(v -> !Double.isFinite(v))) {
  throw new IllegalArgumentException("Lambdas invalid or mismatched with problem");
}

try { double s = problem.check(lambdas); } catch (IllegalStateException e) { /* reload matching model or retrain */ }

When it happens

Trigger: Computing log likelihood after IIS/LCG training when lambdas have drifted to values making the sum s negative — e.g. calling p_iCond / 2LogLikelihood with lambdas read from an incompatible or corrupted model file.

Common situations: Loading lambdas trained on a different problem (different features/empirical distributions); numerical precision drift with extreme lambda values; manual lambda edits.

Understand the failure class

Background: "This is a bug, please report it": internal invariant violations, unreachable panics, and SNH errors explained — this error's family across 47 libraries.

Related errors

• oldTag starts with B, entity at position should not be null
• node cliqueFeatures[n]=
• edge cliqueFeatures[n]=
• : Inconsistent u2b/b2u arrays.
• Suffix pointer moved too far!

AI-assisted analysis of stanfordnlp/CoreNLP@1b7edd19c4 (2026-09-10). Data as JSON: /api/errors/d60f68f7c07b21c4. Report an issue: GitHub.

Appendix: source

Thrown at src/edu/stanford/nlp/maxent/iis/LambdaSolve.java:914

      } //for

    } //for fNo

    for (int x = 0; x < probConds.length; x++) {
      //again
      zlambda[x] = ArrayMath.logSum(probConds[x]); // cpu samples #4,#15: 4.5%
      //log.info("zlambda "+x+" "+zlambda[x]);
      s += zlambda[x] * p.data.ptildeX(x) * p.data.getNumber();

      for (int y = 0; y < probConds[x].length; y++) {
        probConds[x][y] = divide(probConds[x][y], zlambda[x]); // cpu samples #13: 1.6%
        //log.info("prob "+x+" "+y+" "+probConds[x][y]);
      } //y

    }//x

    if (s < 0) {
      throw new IllegalStateException("neg log lik smaller than 0: " + s);
    }

    return s;
  }

  // -- stuff for CG version below -------

  /**
   * calculate the log likelihood from scratch, hashing the conditional
   * probabilities in pcond which we will use for the derivative later.
   *
   * @return The log likelihood of the data
   */
  public double logLikelihoodScratch() {
    // zero all the variables
    double s = 0;
    for (int i = 0; i < probConds.length; i++) {
      for (int j = 0; j < probConds[i].length; j++) {

View on GitHub (pinned to 1b7edd19c4)