# Custom modeling in AWS Clean Rooms ML
<a name="working-with-custom-models"></a>

From a technical standpoint, the following diagram describes how custom ML modeling works in AWS Clean Rooms ML.

![An overview of how AWS Clean Rooms ML works with custom models.](http://docs.aws.amazon.com/clean-rooms/latest/userguide/images/byomHowItWorks.png)


Here's how custom ML modeling works in Clean Rooms ML:

1. Data Source Configuration
   + Source data can be stored in Amazon S3 catalog, in the AWS Glue Data Catalog, or Snowflake
   + AWS Glue Data Catalog is used to organize and catalog
   + Data from multiple AWS accounts can be used within the same collaboration

1. SQL Query and Data Processing
   + SQL queries are used to access and process the source data
   + The queries run within the AWS Clean Rooms collaboration boundaries
   + Processed data feeds into ML Input Channels for model training

1. ML Model Development
   + Source code for the model can be developed using AWS Deep Learning Container Images
   + Custom container images must be created and stored in Amazon Elastic Container Registry

1. Infrastructure Components
   + Amazon Elastic Container Registry stores and manages the ML model containers
   + ML processing occurs within the secure AWS Clean Rooms collaboration environment

1. Monitoring and Logging
   + Amazon CloudWatch provides metrics and logs for both collaborating parties
   + Monitoring is available across AWS accounts involved in the collaboration
   + Performance metrics and operational logs are accessible to relevant parties

1. Results Management
   + Access to results is controlled according to collaboration permissions

Before you get started, see the [Custom ML modeling prerequisites](custom-model-prerequisites.md) and [Model authoring guidelines for the training container](custom-model-guidelines.md) for more information.