Coding the State

Introduction

The State is in a state of development, so this documentation is going to be very minimal.

Configuration

odinson.state in reference.conf contains the config for the state. The StateFactory (ai.lum.odinson.state.StateFactory) uses state.provider to decide what will provide the state. A small list of intended providers is there in the comments. Details needed by each provider are in their own sections, like state.sql. The different providers can optimize for different attributes like speed, persistence, memory usage, etc. and the StateFactory might switch between them depending on runtime conditions.

Operation

Code for the state is located in package ai.lum.odinson.state. Trait State provides the very simple outline. Two methods addResultItems and getResultItems are paired. The third getDocIds is an optimization so that results can be obtained from individual documents in the likely case that retrieving results from all documents is prohibitively expensive.

States are normally transient and only available during a query. They are created in ExtractorEngine.extractMentions by a stateFactory. The state is included in the query calls so that the state can be queried. The query method temporarily sets the state of the OdinsonQuery argument, just before it is used for an odinSearch. If that argument is a composite of additional OdinsonQuery objects, i.e., subqueries, it sets their states as well, recursively. Immediately after the query, the state is cleared. However, the results of odinSearch are added to the state so that subsequent queries that are part of the same extraction process can search the state. The results are of type OdinResults. They are used in the construction of an iterator, an OdinResultsIterator, which is used in State.addResultItems.

A Mention can be created from OdinResults and that is normally what happens at the end of ExtractorEngine.extractMentions. These same OdinResults can be retrieved from the state with State.getResultItems and used to create Mentions as if they had been found somewhere else.