Packages

package engines

Ordering
  1. Alphabetic
Visibility
  1. Public
  2. All

Type Members

  1. sealed trait TextMetric extends AnyRef

Value Members

  1. object DependencyAccuracyEngine

    Backs dependency parsing: UAS (unlabeled attachment score) and LAS (labeled attachment score), computed as a distributed map-then-reduce over per-token (head, label) pairs.

  2. object LabelAccuracyEngine

    Backs the per-item label-accuracy tasks (POS, Classification, SpellCheck, LanguageDetection) and top-k label accuracy (ImageClassification), via Spark's own distributed org.apache.spark.mllib.evaluation.MulticlassMetrics rather than a driver-collected loop, following the precedent in Doc2VecTestSpec.

  3. object SpanF1Engine

    Backs NER and word segmentation: standard conlleval-style entity/segment-level precision/recall/F1, computed as a distributed map (each row to local TP/FP/FN counts) then reduce, rather than collecting tag sequences to the driver.

  4. object TextMetric
  5. object TextSimilarityEngine

    Backs the generated-text tasks (Translation/BLEU, Summarization/ROUGE, SpeechRecognition/WER, QuestionAnswering/SQuAD EM+F1).

    Backs the generated-text tasks (Translation/BLEU, Summarization/ROUGE, SpeechRecognition/WER, QuestionAnswering/SQuAD EM+F1). No JVM equivalents of sacrebleu/rouge_score/jiwer exist, so these are original implementations matching those packages' documented default behavior.

Ungrouped