feat(adr): Add ADR-0011 (Java to Python service migration) #62
Conversation
ruivieira
added
ADR
ruivieira
changed the title
ruivieira
requested review from
AmberJBlue,
RobGeada,
adolfo-ab,
christinaexyou,
dahlem,
danielezonca,
dmaniloff,
m-misiura,
saichandrapandraju,
sheltoncyril and
zanetworker
| - HDF5 for data storage | ||
| - Pandas for data manipulation | ||
| - Prometheus client for metrics exposure | ||
| - MariaDB |
There was a problem hiding this comment.
Hey Rui 👋, quick question, does it have to be MariaDB or could we just say any SQL based DB provider?
There was a problem hiding this comment.
Hi @sheltoncyril , good question.
It does not have to be MariaDB exclusively, but for each supported storage backend we would need to write an interface for it. As a starting point, I think we should do MariaDB first, since that's the supported one for the Java service. But in theory, we could have either TrustyAI (or community) interfaces for any storage type in the future.
| - Service Metadata (`/info/*`) | ||
| - Legacy Endpoints (`/metrics/dir`, `/metrics/spd`) | ||
|
|
||
| **Python components:** |
There was a problem hiding this comment.
Do we need to identify which upstream components provide the implementations that give us parity with the status quo?
There was a problem hiding this comment.
@dahlem good point. IMO the plan is to (later on) replace any algorithm with TrustyAI code/sdk providers.
e.g. we would have in these endpoints something like:
from trustyai.providers import FairnessProvider
# /metrics/spd
spd = FairnessProvider(provider="<your_favourite_library", ...).spd(data)with the concrete metrics backend selected by the user via configuration, but I agree it's a good idea to have specify which will the providers by default (e.g. scikit-learn, AIF360, etc)
3 participants
No description provided.