Sherif Akoush
Sherif Akoush
In `AnchorImage` the channel dimension is assumed to be the last dimension as defined [here](https://github.com/SeldonIO/alibi/blob/ab7dbf9b2a153ff49f29a372d5bdeac56dc7284a/alibi/explainers/anchor_image.py#L235). This is not necessarily true, for example in mnist handwritten pytorch model the shape of...
As part of ongoing work to integrate `alibi` runtime in `mlserver`, we realised that for `IntegratedGradients` if the inference model outputs a class score (e.g mnist model handwritten digit recognition),...
We might also do some validation of the explain parameters that are being passed from config. currently I guess the explainer will throw an error but perhaps we could have...
alibi `detect` is considering init based config or something similar, which will be good to apply for `explain` at some point as well. Currently implementation details of alibi explainers are...
Triton exposes an API to provide details statistics on model usage check [here](https://github.com/triton-inference-server/server/blob/main/docs/protocol/extension_statistics.md): `rpc ModelStatistics(ModelStatisticsRequest)` It would be good to consider something similar for MLServer. Specifically I like `queue` vs...
Currently `TensorDictCodec` is specific to mlflow runtime but it is useful beyond just this runtime. Consider moving it one level up for other runtimes to access it without having to...
If an explainer (e.g. anchors) does not allow batched explanation, then we could in the runtime loop over each element in the batch and conduct explanation of each item. This...
Currently we allow users to pass in runtime explain parameters via setting `explain_parameters` in the `parameters` field of the explanation payload. This allows only basic types. In some cases the...
In some cases we ended up with a experiment db state that contains experiments that fail validation (unclear how yet). This PR skips loading experiments if they fail validation as...
Currently ansible automation does not install grafana. It would be useful to have this as part of the process. Things to consider for configuration: - grafana version - prometheus datasource...