Dynamic Interactions Process

The internal process functions as follows for real-time inference:

• Invocations API is called: When api is called, the engine will have the configuration loaded from startup. On first load certain actions are performed to prepare environment. Housekeeping actions are also performed like creating indexes.
• Setup standard configuration options: All sequence based options that were set in the project deployment will execute for example data lookups, corpora loading.
• Call Pre-Predict Plugin: Pre-score plugin is called to perform any data or eligibility actions. The params JSON Object is carried throughout entire execution process and values can be added for later use.
• Determine algorithm: Each algorithm has a difference treatment process that depends on the type of data needed.
• Generate score via API: Score is generated based on input parameters or training features. Scores are added to params.
• Call Post-Predict Plugin: Post-score plugin is called to perform any data or eligibility actions and access to params allow for full control over all actions.
• Logging: Extensive logging of scores and related items, also logging/ecosystemruntime_history is populated by algorithms that do in-line and cyclical updates. This collection can be used to understand changing scores at a fine-grained leval.
• Payload: Generate payload to return via API.

Use pre- and post-predict logic plugins to extend your use of the overall capability in the platform.

Dynamic Background Process

The background process is controlled by the MONITORING_DELAY setting as outlined in Runtime Settings. This process functions as follows:

• Load configs: Project and Dynamic configurations are loaded on first process cycle, and tested for change every subsequent cycle.
• Process per algorithm: Each algorithm has its own process and uses dynamic config settings and update the options store with latest scores.

Indexes

Also note that this process will create the needed indexes automatically to ensure training and scoring performance.

Logging Collection Indexes

1. uuid index:

{ "uuid" : 1 }

2. date_log index:

{ "date_log" : 1 }

3. stats.predictor, params.value, date_log, and cache index:

{ "stats.predictor" : 1, "params.value" : 1, "date_log" : 1, "cache" : 1 }

4. predictor, params.value, and date_log index:

{ "predictor" : 1, "params.value" : 1, "date_log" : 1 }

Response Collection Indexes

1. uuid index:

{ "uuid" : 1 }

2. date_log, response.offers_accepted.contextual_variable_one, response.offers_accepted.contextual_variable_two, and response.offers_accepted.offer_name index:

{ "date_log" : 1, "response.offers_accepted.contextual_variable_one" : 1, "response.offers_accepted.contextual_variable_two" : 1, "response.offers_accepted.offer_name" : 1 }

3. predictor, date_log, and response.offers_accepted.offer_name index:

{ "predictor" : 1, "date_log" : 1, "response.offers_accepted.offer_name" : 1 }

4. response.offers_accepted index:

{ "response.offers_accepted" : 1 }