The first step of using Mona is to set up the data export from your monitored context.
You can export any JSON serializable data to Mona.
Each JSON you export relates to one specific context instance, identifiable using a context id. You can send several JSONs with complementary information about a single context instance (using the same context id) asynchronously and from different parts of your system. Mona will then automatically aggregate JSON data according to the context id so that you don’t have to worry about getting all the relevant data in one place on your system.
The common approach is to export anything that would be useful in assessing model behavior, including but not limited to meta-data, meta-features, feature vectors, intermediate results, scores, and other outputs.
Mona supports multiple export mechanisms including:
Logging clients: Users can install Mona’s open-source logging client and call it directly from their code. With a single line of code, you can send data to Mona at "model inference time" or elsewhere in your system.
- Python client - see here how to use MONA_SDK
- Additional clients are available upon request (e.g., Java)
REST API: Users who prefer to avoid installing a logging client, can leverage Mona’s REST API endpoint. See here how to format and export your data.
Cloud storage and data lakes: Some users already collect the data they want to monitor in the course of their operations. In these cases, Mona can read data (in various supported formats, e.g., Parquet, JSON) directly from a user’s storage (e.g., Azure Blob, AWS s3 buckets).
Updated about 1 year ago