Najeeb Kazmi

Improve inference performance with loaded TransformerChain ML.NET model

2

Fix #370 PR #230 introduced ability to load and score ML.NET models trained in the new ML.NET `TransformerChain` serialization format. This is done by checking whether "TransformerChain" exists in the...

PR #230 introduced ability to load and score ML.NET models trained in the new ML.NET `TransformerChain` serialization format. This was done by checking whether "TransformerChain" exists in the archive members....

Found while investigating #201 **Describe the bug** Cannot create a `PredictionEngine` in ML.NET from a model trained in NimbusML. Can only use `model.Transform(data)` to make predictions, whether single or batch....

Extra ReLU in L Block

3

https://github.com/openclimatefix/skillful_nowcasting/blob/e2d601a377613b463fcc3c7319017f8051346e8e/dgmr/common.py#L280 I believe this first ReLU should not be present in the L Block, going by the diagram in the [paper](https://arxiv.org/pdf/2104.00954.pdf) (Extended Data Figure 1, page 18).