Coupling (zipping) keys in jgen
conda build 3 comes with a feature that allows to couple keys together: https://conda.io/docs/user-guide/tasks/build-packages/variants.html#coupling-keys
What it does is that it takes several keys and couples them together. It's like calling zip
on the lists.
This reduces the number of combinations that is possible. For example see this:
patch_size: [3, 28]
n_filter_conv1: [8, 32]
n_filter_conv2: [8, 64]
n_units_dense1: [32, 1000]
train_database: ['replay', 'replaymobile', 'oulunpu', 'uvad']
eval_databases:
- ['replay', ['dev', 'eval']]
- ['replaymobile', ['dev', 'eval']]
- ['oulunpu', ['dev', 'eval']]
- ['uvad', ['dev']]
zip_keys:
# model parameters
-
- patch_size
- n_filter_conv1
- n_filter_conv2
- n_units_dense1
What I want here is to actually train 2 models not 2^4
models.
It would be nice to have this in jgen as well.