Hello,
Following scenario: I got a params.yaml which contains an experiment flag “meta.flag”. This parameter should be changeable on the fly and depends on the executed experiment. In my dvc.yaml I specify output files in different stages which refer to the meta.flag when creating the output artifact. However in runtime only the default value of params.yaml is checked and not the provided on the fly parameter. Any idea how to fix this?
Example
params.yaml
meta:
gpu: false
flag: config1
**
dvc.yaml
**
tuning:
cmd: python tuning.py
deps:
- output/preprocess/trainval.parquet
params:
- tuning
outs:
- output/tuning/${meta.flag}_hyperparameter.json
- output/tuning/${meta.flag}_trials.json
command
dvc exp run -f --queue -S meta.flag=abc