easy_rec/python/protos/feature_config.proto

syntax = "proto2";
package protos;

import "easy_rec/python/protos/hyperparams.proto";
import "easy_rec/python/protos/dnn.proto";
import "easy_rec/python/protos/layer.proto";
enum WideOrDeep {
    DEEP = 0;
    WIDE = 1;
    WIDE_AND_DEEP = 2;
}

message AttentionCombiner {
}

message MultiHeadAttentionCombiner {
}

message SequenceCombiner {
    oneof combiner {
        AttentionCombiner attention = 1;
        MultiHeadAttentionCombiner multi_head_attention = 2;
        TextCNN text_cnn = 3;
    }
}

message EVParams {
    optional uint64 filter_freq = 1 [default=0];
    optional uint64 steps_to_live = 2 [default=0];
    // use embedding cache, only for sok hybrid embedding
    optional bool use_cache = 3 [default=false];
    // for sok hybrid key value embedding
    optional uint64 init_capacity = 4 [default=8388608];
    optional uint64 max_capacity = 5 [default=16777216];
}

message FeatureConfig {
    enum FeatureType {
        IdFeature = 0;
        RawFeature = 1;
        TagFeature = 2;
        ComboFeature = 3;
        LookupFeature = 4;
        SequenceFeature = 5;
        ExprFeature = 6;
        PassThroughFeature = 7;
    }

    enum FieldType {
        INT32 = 0;
        INT64 = 1;
        STRING = 2;
        FLOAT = 4;
        DOUBLE = 5;
        BOOL = 6;
    }

    optional string feature_name = 1;

    // input field names: must be included in DatasetConfig.input_fields
    repeated string input_names = 2;
    required FeatureType feature_type = 3 [default = IdFeature];
    optional string embedding_name = 4 [default = ''];
    optional uint32 embedding_dim = 5 [default = 0];

    optional uint64 hash_bucket_size = 6 [default = 0];
    // for categorical_column_with_identity
    optional uint64 num_buckets = 7 [default = 0];

    // only for raw features
    repeated double boundaries = 8;

    // separator with in features
    optional string separator = 9 [default = '|'];

    // delimeter to separator key from value
    optional string kv_separator = 10;

    // delimeter to separate sequence multi-values
    optional string seq_multi_sep = 101;
    // truncate sequence data to max_seq_len
    optional uint32 max_seq_len = 102;

    optional string vocab_file = 11;
    repeated string vocab_list = 12;

    // many other field share this config
    repeated string shared_names = 16;

    // lookup max select element number, default 10
    optional int32 lookup_max_sel_elem_num = 17 [default = 10];

    // max_partitions
    optional int32 max_partitions = 18 [default = 1];

    // combiner
    optional string combiner = 19 [default = 'sum'];

    // embedding initializer
    optional Initializer initializer = 20;

    // number of digits kept after dot in format float/double to string
    // scientific format is not used.
    // in default it is not allowed to convert float/double to string
    optional int32 precision = 21 [default = -1];

    // normalize raw feature to [0-1]
    optional double min_val = 212 [default=0.0];
    optional double max_val = 213 [default=0.0];

    // normalization function for raw features:
    //   such as: tf.math.log1p
    optional string normalizer_fn = 214;

    // raw feature of multiple dimensions
    optional uint32 raw_input_dim = 24 [default=1];

    // sequence feature combiner
    optional SequenceCombiner sequence_combiner = 25;

    // sub feature type for sequence feature
    optional FeatureType sub_feature_type = 26 [default = IdFeature];

    // sequence length
    optional uint32 sequence_length = 27 [default = 1];

    // for expr feature
    optional string expression = 30;

    // embedding variable params
    optional EVParams ev_params = 31;

    // for combo feature:
    //   if not set, use cross_column
    //   otherwise, the input features are first joined
    //     and then passed to categorical_column
    optional string combo_join_sep = 401 [default = ''];
    // separator for each inputs
    // if not set, combo inputs will not be split
    repeated string combo_input_seps = 402;
}

message FeatureConfigV2 {
    repeated FeatureConfig features = 1 ;
    // force place embedding lookup ops on cpu to improve
    // training and inference efficiency.
    optional bool embedding_on_cpu = 2 [default=false];
}

message FeatureGroupConfig {
    optional string group_name = 1;
    repeated string feature_names = 2;

    optional WideOrDeep wide_deep = 3 [default = DEEP];
    repeated SeqAttGroupConfig sequence_features = 4;
    optional bool negative_sampler = 5 [default = false];
}

message SeqAttMap {
    repeated string key = 1;
    repeated string hist_seq = 2;
    repeated string aux_hist_seq = 3;
}

message SeqAttGroupConfig {
    optional string group_name = 1;
    repeated SeqAttMap seq_att_map = 2;
    optional bool tf_summary = 3 [default = false];
    optional DNN seq_dnn = 4;
    optional bool allow_key_search = 5 [default = false];
    optional bool need_key_feature = 6 [default = true];
    optional bool allow_key_transform = 7 [default = false];
    optional bool transform_dnn = 8 [default = false];
}