DVC doesn’t seem to be properly tracking my params.yaml - whenever I run
dvc exp run --set-param model.encoder.kernel_sizes=[7,6,5,4,3,2]
It updates dvc.lock with the new parameters, tells me none of the stages has changed and says it successfully ran the experiment without actually doing anything.
I’ve configured the stage as follows - do I need to explicitly make params.yaml a dependency?
model:
cmd: python src/scripts/train/model.py
deps:
- data/datasets/data.csv
- model/tokenizer.json
- model/embeddings.pt
outs:
- model/model.pt
- model/text_transform.pt
- model/index_to_name.json
params:
- params.yaml:
metrics:
- metrics.json