What causes "Error setting up training" in stanfordnlp/CoreNLP?

Ensure outputPath exists and is writable before training (create the directory, check 'progress' can be written)

stanfordnlp/CoreNLP · error · RuntimeException

Error setting up training

Error message

Error setting up training

What it means

Clusterer.doTraining sets up output writers and loads dev-set documents; any Exception during this setup phase is wrapped in RuntimeException("Error setting up training"). It signals that statistical-coref clusterer training could not even begin (model file, progress file, or training data problems).

Solutions

Ensure outputPath exists and is writable before training (create the directory, check 'progress' can be written)
Verify the dev data path (StatisticalCorefTrainer.setDataPath("dev")) points to a directory with the expected conll/gold files
Check the model file for the classifier is present and loadable
Inspect e.getCause() to see whether it was the writer or the data loading that failed

Example fix

// before
java edu.stanford.nlp.coref.statistical.ClustererTrain  // outputPath=nonexistent/dir
// after
mkdir -p nonexistent/dir   # and ensure dev data exists at configured dataPath
java edu.stanford.nlp.coref.statistical.ClustererTrain

Defensive patterns

Strategy: validation

Validate before calling

java.io.File out = new java.io.File(outputPath);
if (!out.isDirectory() && !out.mkdirs()) throw new IllegalStateException("bad outputPath");
if (!out.canWrite()) throw new IllegalStateException("outputPath not writable");
java.io.File dev = new java.io.File(dataPath, "dev");
if (!dev.isDirectory()) throw new IllegalStateException("dev data missing at " + dev);

Try / catch

try {
  clusterer.doTraining();
} catch (RuntimeException e) {
  if ("Error setting up training".equals(e.getMessage()))
    throw new IllegalStateException("Training setup failed, cause: " + e.getCause(), e);
  throw e;
}

Prevention

Create and permission-check the output directory before long training runs
Verify dev/eval data files exist at the configured data path
Keep the models/dcoref data layout intact
Log e.getCause() to distinguish writer vs data-loading failures

When it happens

Trigger: Running statistical coref training where the output path/model file cannot be opened, the 'progress' writer cannot be created (e.g. unsupported charset or unwritable path), or ClustererDataLoader.loadDocuments fails to read dev data.

Common situations: Missing dev data files at the configured data path; outputPath property pointing to a non-existent directory; running the trainer outside the models/ directory layout it expects.

Related errors

AI-assisted analysis of stanfordnlp/CoreNLP@1b7edd19c4 (2026-09-10). Data as JSON: /api/errors/874c96ea4095fa83. Report an issue: GitHub.

Appendix: source

Thrown at src/edu/stanford/nlp/coref/statistical/Clusterer.java:102

        modelName + "/";
    File outDir = new File(outputPath);
    if (!outDir.exists()) {
      outDir.mkdir();
    }

    PrintWriter progressWriter;
    List<ClustererDoc> trainDocs;
    try {
      PrintWriter configWriter = new PrintWriter(outputPath + "config", "UTF-8");
      configWriter.print(StatisticalCorefTrainer.fieldValues(this));
      configWriter.close();
      progressWriter = new PrintWriter(outputPath + "progress", "UTF-8");

      Redwood.log("scoref.train", "Loading training data");
      StatisticalCorefTrainer.setDataPath("dev");
      trainDocs = ClustererDataLoader.loadDocuments(MAX_DOCS);
    } catch (Exception e) {
      throw new RuntimeException("Error setting up training", e);
    }

    double bestTrainScore = 0;
    List<List<Pair<CandidateAction, CandidateAction>>> examples = new ArrayList<>();
    for (int iteration = 0; iteration < RETRAIN_ITERATIONS; iteration++) {
      Redwood.log("scoref.train", "ITERATION " + iteration);
      classifier.printWeightVector(null);
      Redwood.log("scoref.train", "");
      try {
        classifier.writeWeights(outputPath + "model");
        classifier.printWeightVector(IOUtils.getPrintWriter(outputPath + "weights"));
      } catch (Exception e) {
        throw new RuntimeException();
      }

      long start = System.currentTimeMillis();
      Collections.shuffle(trainDocs, random);

View on GitHub (pinned to 1b7edd19c4)