Class NegativeTrainingDumper

java.lang.Object

org.ek9lang.assist.NegativeTrainingDumper

public class NegativeTrainingDumper
extends Object

Outputs negative training examples as JSONL for LLM fine-tuning. Each NegativeExample produces 1-3 training entries:

1. "What's wrong with this code?" — always generated
2. "How do I fix Exxxxx?" — when howToFix is non-empty
3. "How does EK9 differ from {language}?" — for each language comparison

Output format is mlx_lm chat JSONL with system/user/assistant roles.

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
(package private) static final record	NegativeTrainingDumper.TrainingEntry	A single training entry with a user question and assistant response.

Constructor Summary

Constructors

Constructor	Description
NegativeTrainingDumper()

Method Summary

Modifier and Type	Method	Description
(package private) List<NegativeTrainingDumper.TrainingEntry>	buildTrainingEntries(NegativeExample example)	Build all training entries for a single negative example.
int	writeTrainingData(List<NegativeExample> examples, Path outputDir)	Write negative training data with stratified 90/10 train/valid split by category.

Methods inherited from class Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Details

NegativeTrainingDumper

public NegativeTrainingDumper()

Method Details

writeTrainingData

public int writeTrainingData(List<NegativeExample> examples,
 Path outputDir)
                      throws IOException

Write negative training data with stratified 90/10 train/valid split by category.

Parameters:

examples - the negative examples to output

outputDir - the directory to create (e.g., ./ek9_negative_training_data)

Returns:

the total number of JSONL entries written

Throws:

IOException

buildTrainingEntries

List<NegativeTrainingDumper.TrainingEntry> buildTrainingEntries(NegativeExample example)

Build all training entries for a single negative example.