Add an option to make the aggregate output unique

added 1 commit

• 9c107f73 - keep the whitespace

Compare with previous version

@pkorshunov I added a feature here which makes the lines unique in the aggregate output. This way you can iterate over all combinations but remove redundant lines in the end.

Thank you for this feature. I will check it out.

@pkorshunov I came with a yaml file like this eventually:

# grid_search.yaml
model:
  - {name: lenet_patch_whole_1,
     config: patch_estimator.py,
     patch_size: 28,
     n_filter_conv1: 32,
     n_filter_conv2: 32,
     n_units_dense1: 1000,
     image: whole_image}

train:
  - {database: replaymobile, protocol: grandtest}
  - {database: uvad, protocol: experiment_1}
  - {database: replay, protocol: grandtest}
  - {database: oulunpu, protocol: Protocol_1}

eval:
  - {database: replaymobile, protocol: grandtest, groups: ['dev', 'eval']}
  - {database: uvad, protocol: experiment_1, groups: ['dev', 'eval']}
  - {database: replay, protocol: grandtest, groups: ['dev', 'eval']}
  - {database: oulunpu, protocol: Protocol_1, groups: ['dev']}

and I use it to generate the following template:

# variables.py
class AttrDict(dict):
    __getattr__ = dict.__getitem__
    __setattr__ = dict.__setitem__


model = AttrDict({
    'name': '{{ model.name }}',
    'patch_size': '{{ model.patch_size }}',
})
train = AttrDict({
    'database': '{{ train.database }}',
    'protocol': '{{ train.protocol }}',
})
evaluation = AttrDict({
    'database': '{{ eval.database }}',
    'groups': '{{ eval.groups }}',
    'protocol': '{{ eval.protocol }}',
})

Then in the rest of my config files I can do:

name = model.name
...

assigned to @mguenther

assigned to @andre.anjos

mentioned in commit 43de8af2

merged