Base config

[FynnBe]

base the ModelZooEntry and it's nested configuration elements on a BaseConfig class

Features:

• map to and from a string representation, according to each field's metadata.
• fields can be hidden when representing the ModelZooEntry as a dict structure. Any field that does not require user input (yet) can exist in code, but won't show up in the user friendly config. When loading from such a config a hidden field would have its default value or a value derived from other fields.
  • export the config with yaml (safe for strings/without hidden fields, unsafe for python objects/with hidden fields)

This PR incorporates #1 , #2 , #3 , and #5

[FynnBe]

export of default ModelZooEntry:

:crystal_ball:

entry = ModelZooEntry(map_to=MapTo.STRING)
entry.export(Path("string.yaml"))

log_config:
  model_n_saved: 2
  model_save_interval: 2
loss_config:
  loss_class: MSELoss
  loss_kwargs: {}
model_config:
  model_class: null
  model_kwargs: {}
optimizer_config:
  optimizer_class: Adam
  optimizer_kwargs: {}

:mortar_board:

entry.export(Path("pytorch.yaml"), map_to=MapTo.PYTORCH, export_hidden=True, safe_dump=False)

evaluator_callbacks: []
log_config:
  checkpointer: !!python/object:ignite.handlers.checkpoint.ModelCheckpoint
    _atomic: true
    _dirname: C:/Users/fbeut/protozoo/checkpoints
    _fname_prefix: protozoo
    _iteration: 0
    _n_saved: 2
    _save_as_state_dict: true
    _save_interval: 2
    _saved: []
    _score_function: null
    _score_name: null
  dir: !!python/object/apply:pathlib.WindowsPath
  - C:\
  - Users
  - fbeut
  - protozoo
  model_n_saved: 2
  model_save_interval: 2
loss_config:
  loss_class: !!python/name:torch.nn.modules.loss.MSELoss ''
  loss_kwargs: {}
model_config:
  create_loss_input: !!python/name:protozoo.pytorch_config.default_create_loss_input ''
  create_model_input: !!python/name:protozoo.pytorch_config.default_create_model_input ''
  model_class: null
  model_kwargs: {}
  pretrained_source: null
optimizer_config:
  optimizer_class: !!python/name:torch.optim.adam.Adam ''
  optimizer_kwargs: {}
predictor_callbacks: []
trainer_callbacks: []

[FynnBe]

ok, so somehow these "draft" implementations went way over board... Maybe we'll get back to the details once the higher-level questions are answered and those answers still compatible with what I sketched out here.

[FynnBe]

the current string output:

log_config:
  model_n_saved: 2
  model_save_interval: 2
loss_config:
  loss_class: MSELoss
  loss_kwargs: {}
model_config:
  model_class: NOT SPECIFIED
  model_kwargs: {}
optimizer_config:
  optimizer_class: Adam
  optimizer_kwargs: {}
origin: file:///C:/repos/protozoo/protozoo/entry.py