This change adds an optional APP_EXTRA_URLS configuration section that enables a user to configure more than one app to which to send the input messages. This enables use-cases such as:
1) Deploying a "shadow model" which runs side-by-side with the main model for ongoing evaluation before making the model canonical.
2) Enabling ensembling of models by sending the same data-point to multiple models and merging the resulting predictions in another system.
Note that in order to maintain full backwards compatibility, the input reader currently does not error out if any of the requests to the extra apps failed.
This change adds an optional
APP_EXTRA_URLS
configuration section that enables a user to configure more than one app to which to send the input messages. This enables use-cases such as:1) Deploying a "shadow model" which runs side-by-side with the main model for ongoing evaluation before making the model canonical.
2) Enabling ensembling of models by sending the same data-point to multiple models and merging the resulting predictions in another system.
Note that in order to maintain full backwards compatibility, the input reader currently does not error out if any of the requests to the extra apps failed.