Ground truth for inferences (APIs)

Associates ground truth with model inferences stored the database via the /v1/inference-tracking/ endpoint. When ground truth (i.e., actual values) are available with inferences, you can use performance policies to analyze and track model performance. Examples of ground truth may include true fraud, final loan amount, etc. You can add ground truth to inferences using the /v1/inference-tracking/ground-truth endpoint.

Using preprocessing for model performance

Performance drift policies require ground truth for model inferences. The preprocessing POST endpoint provides a join method to combine the inference table and ground truth table. This must be run before creating policies and segments for a model deployment.

{
    "deployment": "deployment_xyz",
    "model_name": "model_abc",
    "model_stage": "primary",
    "model_version": "1",
    "ground_truths": [{"key": "value", "key": "value"}],
    "format": "file",
    "file_type": "csv",
    "cache_key": "temp.123456",
}

Property	Type	Description
deployment	string	Name of the deployment for these ground truth values.
model_name	string	Name of the model for these ground truth values.
model_version	string	Model version for these ground truth values.
model_stage	string	Model stage for these ground truth values.
ground_truths	list	Ground truth values for the specified columns.
format	string	Format for sending ground truth, as `file` or `raw` (JSON).
file_type	string	If sending as a file, specifies the file type as `csv` or `parquet` Note: The recommended file size is no larger than 500MB. Also, parquet files are recommended over csv files, for reasons of performance (better compression, faster uploads) and data type maintenance.
cache_key	string	If ground truth files are loaded to cache (using the endpoint v1/dataloading/submit), specifies the key for accessing the file in cache.

TABLE OF CONTENTS