What causes "Cannot make normalized counter with Dynamic prior." in stanfordnlp/CoreNLP?

Use a static Distribution as the prior (e.g. Distribution.uniform(counter))

stanfordnlp/CoreNLP · error · UnsupportedOperationException

Cannot make normalized counter with Dynamic prior.

Error message

Cannot make normalized counter with Dynamic prior.

What it means

distributionWithDirichletPrior() mixes a counter with a prior Distribution, but a DynamicDistribution prior can change over time and therefore cannot yield a fixed normalized mixture. The method detects this and throws UnsupportedOperationException rather than producing an incorrect static distribution.

Solutions

Use a static Distribution as the prior (e.g. Distribution.uniform(counter))
Materialize the dynamic prior into a static snapshot Distribution before mixing
If dynamic behavior is required, implement the mixture manually and recompute on each prior update

Example fix

// before
Distribution<String> d = Distribution.distributionWithDirichletPrior(c, dynamicPrior, 1.0);
// after
Distribution<String> staticPrior = Distribution.uniform(c);
Distribution<String> d = Distribution.distributionWithDirichletPrior(c, staticPrior, 1.0);

Defensive patterns

Strategy: type-guard

Validate before calling

if (prior instanceof DynamicDistribution) {
  // choose a static prior instead
}

Type guard

boolean isStaticPrior(Distribution<?> prior) { return !(prior instanceof DynamicDistribution); }

Try / catch

try {
  Distribution<E> d = Distribution.distributionWithDirichletPrior(c, prior, weight);
} catch (UnsupportedOperationException e) {
  prior = Distribution.uniform(c);
  Distribution<E> d = Distribution.distributionWithDirichletPrior(c, prior, weight);
}

Prevention

Keep dynamic priors out of static-mixing APIs; snapshot them first
Type-check the prior with instanceof DynamicDistribution at call sites
Document prior requirements in wrapper methods

When it happens

Trigger: Calling Distribution.distributionWithDirichletPrior(c, prior, weight) where prior is an instance of DynamicDistribution (or a subclass).

Common situations: Developers switching from a static prior (e.g. Distribution.uniform) to an adaptive/online prior without realizing the Dirichlet-mixing path does not support dynamic priors.

Understand the failure class

Background: UnsupportedOperationException and "is not supported" errors: when a library deliberately refuses a call — this error's family across 30 libraries.

Related errors

AI-assisted analysis of stanfordnlp/CoreNLP@1b7edd19c4 (2026-09-10). Data as JSON: /api/errors/d91adc6480b92790. Report an issue: GitHub.

Appendix: source

Thrown at src/edu/stanford/nlp/stats/Distribution.java:548

  /**
   * Returns a Distribution that uses prior as a Dirichlet prior
   * weighted by weight.  Essentially adds "pseudo-counts" for each Object
   * in prior equal to that Object's mass in prior times weight,
   * then normalizes.
   * <p>
   * WARNING: If unseen item is encountered in c, total may not be 1.
   * NOTE: This will not work if prior is a DynamicDistribution
   * to fix this, you could add a CounterView to Distribution and use that
   * in the linearCombination call below
   *
   * @param weight multiplier of prior to get "pseudo-count"
   * @return new Distribution
   */
  public static <E> Distribution<E> distributionWithDirichletPrior(Counter<E> c, Distribution<E> prior, double weight) {
    Distribution<E> norm = new Distribution<>();
    double totalWeight = c.totalCount() + weight;
    if (prior instanceof DynamicDistribution) {
      throw new UnsupportedOperationException("Cannot make normalized counter with Dynamic prior.");
    }
    norm.counter = Counters.linearCombination(c, 1 / totalWeight, prior.counter, weight / totalWeight);
    norm.numberOfKeys = prior.numberOfKeys;
    norm.reservedMass = prior.reservedMass * weight / totalWeight;
    //System.out.println("totalCount: " + norm.totalCount());
    return norm;
  }

  /**
   * Like normalizedCounterWithDirichletPrior except probabilities are
   * computed dynamically from the counter and prior instead of all at once up front.
   * The main advantage of this is if you are making many distributions from relatively
   * sparse counters using the same relatively dense prior, the prior is only represented
   * once, for major memory savings.
   *
   * @param weight multiplier of prior to get "pseudo-count"
   * @return new Distribution
   */

View on GitHub (pinned to 1b7edd19c4)