sklearn

Description

We want to add support for RNNs. I will let this issue open for people who want to contribute to this project.
My suggestion would be just to write RNN (or any other RNN-like model) in the algorithm field in order to use an RNN model. Here is an example to illustrate what I mean:

model:
     type: classification
     algorithm: RNN
    .
    . 

**If you a

/kind feature

Describe the solution you'd like

In pkg/apis/serving/v1beta1/inference_service_defaults.go the default InferenceService resource requests and limits are hard coded to be 1 cpu and 2Gi memory. These are reasonable defaults. However, the entire existence of these defaults should be disablable. Moreover, administrators should be able to quickly adjust defaults globally via t