tribler.core.database.augmenter
===============================

.. py:module:: tribler.core.database.augmenter


Attributes
----------

.. autoapisummary::

   tribler.core.database.augmenter.logger
   tribler.core.database.augmenter.LOG_LEVELS


Classes
-------

.. autoapisummary::

   tribler.core.database.augmenter.AugmentedSearch


Module Contents
---------------

.. py:data:: logger

.. py:data:: LOG_LEVELS

   logging              {50: CRITICAL/FATAL, 40: ERROR, 30: WARNING/WARN, 20: INFO, 10: DEBUG, 0: NOTSET}
   SentencePieceTrainer { 3:          FATAL,  2: ERROR,  1:      WARNING,  0: INFO                      }


.. py:class:: AugmentedSearch(config: tribler.tribler_config.TriblerConfigManager, notifier: tribler.core.notifier.Notifier, task_manager: ipv8.taskmanager.TaskManager)

   This class is responsible for "slow" search. It creates its own language, based on torrent titles, to augment
   user queries.


   .. py:attribute:: config


   .. py:attribute:: max_title_length
      :value: 4192



   .. py:attribute:: title_window
      :type:  list[str]
      :value: []



   .. py:attribute:: initialized
      :value: False



   .. py:attribute:: write_completed
      :type:  asyncio.Future[None]


   .. py:attribute:: task_manager


   .. py:attribute:: model_file


   .. py:attribute:: title_cache_file


   .. py:attribute:: processor
      :type:  sentencepiece.SentencePieceProcessor


   .. py:method:: on_shutdown() -> None

      Quickly dump our cache onto the disk.



   .. py:method:: consume_torrent_metadata(metadata: dict) -> None

      We found a new torrent title. Put it in the cache and start learning when we have sufficient data.



   .. py:method:: write(trained_model: bytes) -> None

      SentencePiece really wants to write to disk and then load it again. We need to be sure that new input was
      actually written to disk before reading it, or we lose data.



   .. py:method:: study() -> None
      :async:


      Study our most recent torrent titles and add them to the top tokens that we have already found in the past.



   .. py:method:: needs_kickstart() -> bool

      Check if we could use a good seeding.



   .. py:method:: to_phrases(pieces: list[str]) -> list[list[str]]

      What we get from SentencePiece is pieces. What we want is the pieces that make up a word, i.e., a phrase.
      Note that each word is separated by "▁" and may span several list elements.



   .. py:method:: augment(search: str, limit: int = 1000, offset: int = 1) -> tuple[str, list[str]]

      Augment the original user search string and create an SQL-injection-safe SQL query and its parameters.