The Mooclet Service integration is too much to maintain when if all we're going to use it for is relatively straightforward thompson-sampling variations. Originally it seemed to be a requirement of the EASI grant, and at that time it seemed we were excited to be able to use a lot of different advanced algorithms.
But it is not an enterprise-grade application, it's undocumented and buggy, and the extra infrastructure is costly, it has many aging components that require dependency updates that are going to get harder to justify fixing, such as outdated Python and Django versions that are becoming obsolete.
We can re-implement a ts_configurable-like native implementation and be rid of the mooclet stuff easily by fall.
The Mooclet Service integration is too much to maintain when if all we're going to use it for is relatively straightforward thompson-sampling variations. Originally it seemed to be a requirement of the EASI grant, and at that time it seemed we were excited to be able to use a lot of different advanced algorithms.
But it is not an enterprise-grade application, it's undocumented and buggy, and the extra infrastructure is costly, it has many aging components that require dependency updates that are going to get harder to justify fixing, such as outdated Python and Django versions that are becoming obsolete.
We can re-implement a ts_configurable-like native implementation and be rid of the mooclet stuff easily by fall.