### The whole config is loaded into src/utils/config.py Config dataclass.
experiment:
  directory: dr_strong # Identifier, used for storing the results.
  n_iter: 500 # Number of repeated independent runs.
  ablation: # Ablation object that is loaded by `sklearn.model_selection.ParameterGrid`. Check it for a detailed explanation of possible types.
    dataset.kwargs["dataset"]: [ecoli, glass, letter, optdigits, page-blocks, pendigits, satimage, vehicle, yeast] # This will change the specified key in the config for every value in the list. You can specify multiple keys at once, see `configs/tuning.yaml`.
dataset:
  module: openml # Name of the file under src/datasets to load the data loading function.
  kwargs: # Hyper-parameters passed to the data loading function.
    dataset: ecoli # This value is completely ignored as the ablation changes it.
    version: 1
  preprocessing: # Preprocessing functions called on the dataset before it is passed to policies.
    - name: ordinal_label # Preprocessing function identifier from datasets/preprocessing.py.
  eval_model_split: 0.5 # Split ratio where `eval` is the set for evaluation and `model` is the set for policy learning.
logging_policy:
  name: softmax # Policy identifier from src/utils/policy.py get_policy function.
  kwargs: # Policy function arguments
    temperature: 1
    model_kwargs:
      max_iter: 1000000
    preprocessing: # Preprocessing functions called specifically before this policy learning.
      - name: subsample
        kwargs:
          n: 1.0
          replace: True
target_policy:
  name: softmax
  kwargs:
    temperature: 10
    model_kwargs:
      max_iter: 1000000
    preprocessing:
      - name: subsample
        kwargs:
          n: 1.0
          replace: True
estimators: # List of estimators to evaluate.
  - name: IPS # The name identifies the class to load as defined in src/estimators/__init__.py file.
  - name: DM
  - name: DR
  - name: EstimatorSelection
    printname: \ensuremath{\tt OCV_{IPS}} # The name this estimator has in legend when plotted.
    hyperparameters: # Hyper-parameter grid for this estimator loaded by `sklearn.model_selection.ParameterGrid`. Can have multiple keys specified. Each combination is then passed to the method as its **kwargs.
      - estimator: [IPS, DM, DR]
    tuning: # If no tuning method is provided, the estimator is evaluated for all hyper-parameter combinations from its grid and all results are logged.
      tuner: OCV # Tuning method that gets imported from the src/tuning/__init__.py file.
      kwargs: # Tuning method hyper-parameters.
        valid_estimator: IPS
  - name: EstimatorSelection
    printname: \ensuremath{\tt OCV_{DR}}
    hyperparameters:
      - estimator: [IPS, DM, DR]
    tuning:
      tuner: OCV
      kwargs:
        valid_estimator: DR
  - name: EstimatorSelection
    printname: \textsc{Slope}
    hyperparameters:
      - estimator: [IPS, DR, DM]
    tuning:
      tuner: SLOPE
  - name: EstimatorSelection
    printname: \ensuremath{\tt PAS{\text -}IF}
    hyperparameters:
      - estimator: [IPS, DM, DR]
    tuning:
      tuner: PASIF
plot:
  type: catplot # Type of plot to use, identifiers are defined in src/utils/store_results.py file. 
  xlabel: Dataset
  use_latex: False # Whether to use LaTeX in matplotlib for printed texts. See https://matplotlib.org/stable/users/explain/text/usetex.html for more details.

### The complete list of configurable parameters with their default values can be found in the src/utils/config.py file.