mindspore.experimental.es.EsEmbeddingLookup

View Source On Gitee
class mindspore.experimental.es.EsEmbeddingLookup(table_id, es_initializer, embedding_dim, max_key_num, optimizer_mode=None, optimizer_params=None, es_filter=None, es_padding_key=None, es_completion_key=None)[source]

Look up a PS embedding.

Warning

This is an experimental EmbeddingService API that is subject to change.

Parameters

  • table_id (int) – The table id.

  • es_initializer (EsInitializer) – The EsInitialize object for PS embedding with table_id, which can be None when the inference is performed.

  • embedding_dim (int) – The embedding dim of keys for PS embedding with table_id.

  • max_key_num (int) – The num of keys when lookup.

  • optimizer_mode (str) – The type of optimizer. Default is None.

  • optimizer_params (tuple[float]) – The parameters of optimizer. Default is None.

  • es_filter (CounterFilter) – The option of counter filter for PS embedding with table_id. Default is None.

  • es_padding_key (PaddingParamsOption) – The option of padding key for PS embedding with table_id. Default is None.

  • es_completion_key (CompletionKeyOption) – The option of completion key for PS embedding with table_id. Default is None.

Inputs:

  • keys (Tensor): The keys of each feature in PS embedding.

  • actual_keys_input (Tensor): Tensor composed of all unique elements of keys.

  • unique_indices (Tensor): The index value of each element in keys to actual_keys_input .

  • key_count (Tensor): The count of each element in the actual_keys_input to keys.

Supported Platforms:

Atlas A2 training series products