Zingg matching failed error "training data has no positive examples labeled"

TL;DR

Label positives before training: Zingg needs labelled positive pairs to learn from, so run the labelling phase, mark enough confirmed matches, and feed that training data back before expecting the model to train. Zingg learns match weights from examples. With no positive examples in the training data, there is nothing to learn the shape of a match from, and training stops with this error.

Zingg matching failed error "training data has no positive examples labeled"

Use this when

• The check or run fails with this symptom: a Zingg matching fails with training data has no positive examples labeled.
• You are working on a Zingg entity matching job being trained on a new dataset and the failure repeats on the same data or configuration.
• You need the failing rows, rule, or stage isolated before changing thresholds or code.

Not for this skill when

• A dataset where positives were labelled but under a different configuration or field mapping, which needs the training data re-linked rather than new labelling.
• The error text names a different tool, stage, or root cause. Match the verbatim error first and use that skill.

Steps

1. Open the training data Zingg generated and count the labelled pairs by label. Verify: positives are zero or far below the other labels
2. Run the Zingg labelling phase and work through the presented pairs, marking genuine matches. Verify: a set of confirmed positive pairs exists
3. Include hard positives, pairs that match despite name or address differences. Verify: the positives cover the variation the model must learn
4. Retrain with the labelled data. Verify: training completes and produces a model
5. Evaluate the model on a held out labelled set. Verify: match quality is measurable before the full run

Variant phrasings

zingg training data has no positive examples labeled

The shorter phrasing focused on the named failure.

zingg matching failed no positives error

The symptom phrasing a searcher tries next.

zingg labelling positive examples fix

The variant naming the failing stage or tool directly.

Compatibility: Zingg training and labelling phases, entity matching workflows

Why it happens

Active learning in Zingg starts from pairs the system proposes, and a labelling session that only marked negatives, or was skipped entirely, leaves the learner with no example of a match. The error is literal: the positive class is empty.

Edge cases / pitfalls

• A handful of positives is enough to start but produces a weak model; label in rounds and retrain between them.
• Positives labelled under an old schema must be remapped when fields change, or training reads them as a different dataset.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst__fP9hJSmNxuM6eK35-cm6A