toyML

介绍

工业界常见机器学习算法，实现：

• Learning to Rank

  • LambdaGBDT
  • LambdaDNN

• CTR预估

  • LR (Logistic Regression)
  • XGBoost + LR
  • DIN (Deep Interest Network)

• Re-ranking 重排序

  • LambdaDNN
  • DLCM (Deep Listwise Context Model)
  • PRM (Personalized Re-ranking Model)

• 召回

  • itemCF
  • node2vec
  • YouTubeDNN
  • SBCNM (Sampling-Bias-Corrected Neural Model)
  • DSSM (Deep Structured Semantic Model)

Hive-UDF实现tf.feature_column特征处理，包括：

• 离线特征哈希，categorical_column_with_hash_bucket
• 离散特征词典索引，categorical_column_with_vocabulary_list
• 连续特征分箱，bucketized_column