[Python知识库] 基于tf1.x的feature_columns相关特征处理方法到tf2.x时的迁移

下表是对应的特征处理表逻辑：keras layer中对应的处理类更加直接易于理解。?

例如连续性数字的离散化处理layers.Discretization

字符型或无大小关系的hash桶化：layers.Hashing

字符串的数字化编码处理：layers.StringLookup

数字类型（无大小区分）数据的数字化编码：layers.IntegerLookup

上面三个可以将输入统一化为one_hot、multi_hot或integer

对于integer形式，可以再通过使用layers.CategoryEncoding实现指定维度的映射变换（可训练）

Feature Column	Keras Layer
feature_column.bucketized_column	layers.Discretization
feature_column.categorical_column_with_hash_bucket	layers.Hashing
feature_column.categorical_column_with_identity	layers.CategoryEncoding
feature_column.categorical_column_with_vocabulary_file	layers.StringLookup?or?layers.IntegerLookup
feature_column.categorical_column_with_vocabulary_list	layers.StringLookup?or?layers.IntegerLookup
feature_column.crossed_column	layers.experimental.preprocessing.HashedCrossing
feature_column.embedding_column	layers.Embedding
feature_column.indicator_column	`output_mode='one_hot'`?or?`output_mode='multi_hot'`*
feature_column.numeric_column	layers.Normalization
feature_column.sequence_categorical_column_with_hash_bucket	layers.Hashing
feature_column.sequence_categorical_column_with_identity	layers.CategoryEncoding
feature_column.sequence_categorical_column_with_vocabulary_file	layers.StringLookup,?layers.IntegerLookup, or?`layer.TextVectorization`?
feature_column.sequence_categorical_column_with_vocabulary_list	layers.StringLookup,?layers.IntegerLookup, or?`layer.TextVectorization`?
feature_column.sequence_numeric_column	layers.Normalization
feature_column.weighted_categorical_column	layers.CategoryEncoding