stanfordnlp/CoreNLP · error · IllegalArgumentException

Cannot handle weird double: " + d

Error message

Cannot handle weird double: " + d

What it means

SloppyMath.segmentDouble() decomposes a double into a mantissa/exponent pair, which is only meaningful for finite numbers. The library throws IllegalArgumentException when the input is infinite or NaN because such values have no finite mantissa/exponent representation.

Solutions

1. Check the value with Double.isFinite(d) before calling segmentDouble and handle non-finite input separately
2. Trace where the NaN/Infinity originated (usually a division by zero or overflow earlier in the pipeline) and fix the upstream computation
3. If non-finite values are expected, clamp or sanitize them before segmenting

Example fix

// before
Triple<Boolean, Long, Integer> t = SloppyMath.segmentDouble(score);
// after
if (!Double.isFinite(score)) {
  score = 0.0; // or handle/log the bad value
}
Triple<Boolean, Long, Integer> t = SloppyMath.segmentDouble(score);

if (!Double.isFinite(d)) { throw new IllegalArgumentException("Expected finite double, got: " + d); }

boolean isUsableDouble(Object v) { return v instanceof Double && Double.isFinite((Double) v); }

try { segmentDouble(d); } catch (IllegalArgumentException e) { /* sanitize d and retry or skip */ }

When it happens

Trigger: Calling SloppyMath.segmentDouble(Double.NaN), segmentDouble(Double.POSITIVE_INFINITY), or segmentDouble(Double.NEGATIVE_INFINITY). Also triggered indirectly when an upstream computation overflows to Infinity or produces NaN (e.g. 0.0/0.0) and the result is passed in.

Common situations: Numerical overflow in probability/log computations in NLP pipelines; dividing by zero-size counts; uninitialized fields that default to NaN being passed to math helpers.

Understand the failure class

Background: "Must be a positive integer", "Invalid value", "Unsupported": the invalid-argument-value error family, when a library rejects the value you pass — this error's family across 35 libraries.

Related errors

• Can't normalize an array with sum 0.0 or NaN: " +…
• Can't normalize an array with sum 0.0 or NaN
• Can't normalize an array with sum 0.0 or NaN: " +…
• Can't sample from NaN
• Can't standardize array whose mean is NaN

AI-assisted analysis of stanfordnlp/CoreNLP@1b7edd19c4 (2026-09-10). Data as JSON: /api/errors/52739ac9db1544bc. Report an issue: GitHub.

Appendix: source

Thrown at src/edu/stanford/nlp/math/SloppyMath.java:709

  /**
   * Taken from http://nerds-central.blogspot.com/2011/05/high-speed-parse-double-for-jvm.html
   */
  public static double parseDouble(boolean negative, long mantissa, int  exponent) {
     // Do this with no locals other than the arguments to make it stupid easy
     // for the JIT compiler to inline the code.
    int e = -16;
    return (negative ? -1. : 1.) * (((double)mantissa) * exps[(e + 308)]) * exps[(exponent + 308)];
  }


  /**
   * Segment a double into a mantissa and exponent.
   */
  public static Triple<Boolean, Long, Integer> segmentDouble(double d) {
    if (Double.isInfinite(d) || Double.isNaN(d)) {
      throw new IllegalArgumentException("Cannot handle weird double: " + d);
    }
    boolean negative = d < 0;
    d = Math.abs(d);
    int exponent = 0;
    while (d >= 10.0) {
      exponent += 1;
      d = d / 10.;
    }
    while (d < 1.0) {
      exponent -= 1;
      d = d * 10.;
    }
    return Triple.makeTriple(negative, (long) (d * 10000000000000000.), exponent);
  }


  /**
   * From http://nadeausoftware.com/articles/2009/08/java_tip_how_parse_integers_quickly

View on GitHub (pinned to 1b7edd19c4)