[IGNITE-11655] [ML]: OneHotEncoder returns more columns than expected

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 2.7
• Fix Version/s: 2.8
• Component/s: ml
• Release Note: OneHotEncoder returns more columns than expected
• Ignite Flags: Release Notes Required
• Assignee: Alexey Zinoviev
• Reporter: Anton Dmitriev
• Created: 29/Mar/19 09:42
• Updated: 14/Jan/20 09:05
• Resolved: 29/Mar/19 15:27

Description

OneHotEncoder returns more columns than expected (two values that might be encoded using two columns encoded using 3 columns). The following example demonstrates the problem:

Map<Integer, Object[]> training = new HashMap<>();

training.put(0, new Object[]{42.0});
training.put(1, new Object[]{43.0});
training.put(2, new Object[]{42.0});

EncoderTrainer<Integer, Object[]> trainer = new EncoderTrainer<Integer, Object[]>()
    .withEncoderType(EncoderType.ONE_HOT_ENCODER)
    .withEncodedFeature(0);

IgniteBiFunction<Integer, Object[], Vector> processor = trainer.fit(training, 1, (k, v) -> v);
Vector res = processor.apply(1, new Object[]{42.0});

System.out.println(Arrays.toString(res.asArray()));

>>> [0.0, 1.0, 0.0]

Issue Links

• links to: GitHub Pull Request #6376
• links to: TC Build