The model-configuration context is currently quite convoluted:
It uses GET /as-configured-model to convert a new model-config into the AMR format
Operates on the AMR
It returns to us the AMR, which we then call POST /as-configured-model to convert it back to the model-configuration format in Terarium
The major issues are:
The conversion is lossy, we lose information converting to and back to Terarium's configuration
Lack support to handle distributions
Lack support to changes things en mass (e.g. when dealing with stratified model configuration)
Task
We should just support our data format directly rather than doing this round-about route where we lose data along the way. We should modify the model-configuration context.
Just pass in the configuration JSON directly instead of an id
Keep an in-memory copy of the JSON and used it for the modify operations
Add support to change data types of parameters, to support both constant and distributions. Note that the distribution need to match the distributions supported in MIRA, as agreed upon with other performers.
Nice to have, maybe 2nd round
Allow changing values en mass, e.g. for configuring stratified models
Note this is a beaker task.
Preamble
The model-configuration context is currently quite convoluted:
GET /as-configured-modelto convert a new model-config into the AMR formatPOST /as-configured-modelto convert it back to the model-configuration format in TerariumThe major issues are:
Task
We should just support our data format directly rather than doing this round-about route where we lose data along the way. We should modify the model-configuration context.
Nice to have, maybe 2nd round