--embedding_keep_msg

Switch

--embedding_keep_msg

Description

Stores the embeddings as a message-level feature table (grouped by --messageid_field) instead of the usual group-level aggregation, when extracting features with --bert_model/--emb_model. Aliased as --emb_keep_msg.

Argument and Default Value

No argument. Boolean switch, default: False.

Details

Known issue — broken in SQLite mode. With this switch, FeatureExtractor.addEmbTable() writes rows via mm.executeWriteMany(...) using SQL built with MySQL's %s parameter placeholders — SQLite's driver expects ? instead. Verified: fails with sqlite3.OperationalError: near "%": syntax error right at the insert step, after the destination table (feat$...$<corptable>$message_id) is already created empty. This is the same placeholder-style bug confirmed for --predict_cv_to_feats. The default (group-level) path does not hit this — confirmed working without this flag. Works against MySQL.

Other Switches

Required Switches:

• --bert_model or --emb_model

Example Commands

Requires MySQL — broken against SQLite, see Known issue above:

dlatkInterface.py -d dla_tutorial -t msgs -c user_id --add_bert --embedding_keep_msg