Collectors
Capture inputs used for making predictions in your app
Collector is an ingenious feature that allows you to capture the inputs used for making predictions. After creating a collector, which includes specifying the caller user ID and the source model, a triggering process is established.
This process ensures that whenever the stated user makes a prediction using the specified model, the inputs used in generating the predictions are automatically ingested and stored in your app.
With collectors, you can automatically pipe in data from production models, gather inputs to feed your models with real-world training data, and unlock many platform training capabilities.
You can learn how to create collectors via the UI here.
The Collector feature is currently exclusively available to our Professional and Enterprise users. Learn more here.
Collector Parameters
Let's talk about the parameters required to create a collector via the API.
Collector ID
Give your collector a useful and descriptive name.
Description
Provide additional details about your collector.
Pre-queue Workflow
In many scenarios, you will only want to ingest a sample, or subset of a given data source into your app. Pre-queue workflows allow you to pre-process your inputs so that you can sample and filter your new data before it is ever added to your app. Pre-queue workflows allow you to specify sampling rules for triggering data ingestion.
Common pre-queue workflows are designed to:
- Randomly sample inputs
- Filter inputs by metadata
- Filter inputs with a maximum probability below a given threshold
- Filter inputs with a minimum probability above a given threshold
- Filter specific concept probabilities above a given threshold
- Undertake knowledge graph mapping from public General model concepts to a custom model
At least one (pre-queue or post-queue) workflow ID is required.