config.yml.example

# --------------------------------------------------------------------------
# Example Configuration for the Ensemble Genetic Algorithm Project
#
# To use, copy this file to `config.yml` and uncomment/edit the parameters
# you wish to change. Any parameter not specified in `config.yml` will use
# its hardcoded default value, which is shown here for reference.
# --------------------------------------------------------------------------
 
# --- Global Parameters ---
# These settings control the overall behavior of the experiment.
# Corresponds to attributes in `ml_grid.util.global_params.global_parameters`.
global_params:
  # Path to the input dataset CSV file.
  # Default: "synthetic_data_for_testing.csv"
  # input_csv_path: "path/to/your/data.csv"

  # The total number of grid search iterations to perform.
  # Default: 1
  # n_iter: 50 # For a full experiment, a higher number is recommended.

  # If True, enables testing mode (e.g., smaller data samples).
  # Default: True
  # testing: False

  # Number of samples to use for the test set.
  # Default: 500
  # test_sample_n: 500

  # Number of columns to sample from the data (for testing). 0 means all columns.
  # Default: 30
  # column_sample_n: 30

  # List of base learner models to include in the search space.
  # Default: A list of all available models.
  # model_list:
  #   - "logisticRegression"
  #   - "perceptron"
  #   - "extraTrees"
  #   - "randomForest"
  #   - "kNearestNeighbors"
  #   - "XGBoost"
  #   - "DecisionTreeClassifier"
  #   - "AdaBoostClassifier"
  #   - "elasticNeuralNetwork"
  #   - "GaussianNB"
  #   - "QuadraticDiscriminantAnalysis"
  #   - "SVC"
  #   - "DummyModel" # For testing purposes

  # Verbosity level for console output (0-9).
  # Default: 3
  # verbose: 3
 
  # Debug level for detailed logging.
  # Default: 0
  # debug_level: 0
 
  # Number of parallel jobs for KNN models. -1 uses all available processors.
  # Default: -1
  # knn_n_jobs: -1
 
  # If True, sanitizes column names for compatibility with libraries like XGBoost.
  # Default: True
  # rename_cols: True
 
  # If True, raises exceptions during grid search; otherwise, logs them.
  # Default: True
  # error_raise: True
 
  # If True, uses RandomizedSearchCV instead of GridSearchCV.
  # Default: True
  # random_grid_search: True
 
  # Percentage of the parameter space to sample in a random grid search.
  # Default: 0.001
  # sub_sample_param_space_pct: 0.001
 
  # Number of parallel jobs for grid search. -1 uses all available cores.
  # Default: 4
  # grid_n_jobs: 4
 
  # Time in seconds after which a warning is printed for long model training times.
  # Default: 60
  # model_train_time_warning_threshold: 60
 
  # If True, saves trained base learners to disk. Can use a lot of space.
  # Note: This is a global default. The `store_base_learners` setting
  # in `grid_params` will override this for each individual grid search run.
  # Default: False
  # store_base_learners: False
 
  # Number of generations without improvement before the GA stops early.
  # Default: 5
  # gen_eval_score_threshold_early_stopping: 5
 
  # The base name for the log file that stores experiment results.
  # Default: "log_store_dataframe"
  # log_store_dataframe_path: "log_store_dataframe"
 
  # The root directory for saving project outputs (logs, models, etc.).
  # Default: "HFE_GA_experiments"
  # base_project_dir: "HFE_GA_experiments"

# --- Genetic Algorithm Parameters ---
# These control the evolution process itself (population size, generations, etc.).
# Corresponds to attributes in `ml_grid.util.grid_param_space_ga.Grid`.
# Note: These values are overridden if `global_params.testing` is set to `True`.
ga_params:
  # List of possible values for the number of base learners in an ensemble.
  # Default: [4, 8, 16, 32, 64]
  # nb_params: [4, 8, 16, 32, 64]
 
  # List of possible values for the population size.
  # Default: [32, 64, 128]
  # pop_params: [32, 64, 128]
 
  # List of possible values for the number of generations.
  # Default: [128]
  # g_params: [128]
 
# --- Grid Search Hyperparameters ---
# This section defines the search space for each experiment iteration.
# Corresponds to the `grid` dictionary in `ml_grid.util.grid_param_space_ga.Grid`.
grid_params:
  # --- Ensemble & Preprocessing Parameters ---
  # Methods for weighting base learner predictions. Options: "ann", "de", "unweighted".
  # Default: ["ann", "de", "unweighted"]
  # weighted: ["ann", "de", "unweighted"]
 
  # Whether to reuse stored base learners from previous runs.
  # Default: [False]
  # use_stored_base_learners: [False]
 
  # Whether to store newly trained base learners to disk for this run.
  # Default: [False]
  # store_base_learners: [False]
 
  # Resampling methods for imbalanced data. Options: "undersample", "oversample", None.
  # Default: ["undersample", "oversample", None]
  # resample: ["undersample", "oversample", None]
 
  # Whether to scale features.
  # Default: [True]
  # scale: [True]
 
  # Number of features to use.
  # Default: ["all"]
  # n_features: ["all"]
 
  # Size of hyperparameter space for base learners.
  # Default: ["medium"]
  # param_space_size: ["medium"]
 
  # Number of unique outcomes.
  # Default: [10]
  # n_unique_out: [10]
 
  # Outcome variable identifier (e.g., "outcome_var_1").
  # Default: ["1"]
  # outcome_var_n: ["1"]
 
  # Diversity penalty factor for ensemble evaluation.
  # Default: [0]
  # div_p: [0]
 
  # --- Feature Selection Parameters ---
  # Correlation threshold for removing columns.
  # Default: [0.9, 0.99]
  # corr: [0.9, 0.99]
 
  # Percentage of missing data allowed in a column before it's removed.
  # Default: [99.9, 99.8, 99.7]
  # percent_missing: [99.9, 99.8, 99.7]
 
  # Method for feature selection.
  # Default: ["anova"]
  # feature_selection_method: ["anova"]
 
  # --- Genetic Algorithm Hyperparameters (per grid search run) ---
  # Crossover probability.
  # Default: [0.5, 0.75, 0.25]
  # cxpb: [0.5, 0.75, 0.25]
 
  # Mutation probability.
  # Default: [0.2, 0.4, 0.8]
  # mutpb: [0.2, 0.4, 0.8]
 
  # Probability of individual gene mutation.
  # Default: [0.025, 0.05, 0.075]
  # indpb: [0.025, 0.05, 0.075]
 
  # Tournament size for selection.
  # Default: [3, 6, 9]
  # t_size: [3, 6, 9]
 
  # --- Data Feature Selection ---
  # This nested dictionary controls which groups of features are included in the search space.
  # The default for each feature group is a list of booleans, e.g., [True, False],
  # meaning the grid search will test both including and excluding that group.
  # To force inclusion, use [True]. To force exclusion, use [False].
  # data: [
  #   {
  #     age: [True],
  #     sex: [True],
  #     bmi: [True],
  #     ethnicity: [True],
  #     bloods: [True, False],
  #     diagnostic_order: [True, False],
  #     drug_order: [True, False],
  #     annotation_n: [True, False],
  #     meta_sp_annotation_n: [True, False],
  #     annotation_mrc_n: [True, False],
  #     meta_sp_annotation_mrc_n: [True, False],
  #     core_02: [True],
  #     bed: [True],
  #     vte_status: [False],
  #     hosp_site: [False],
  #     core_resus: [False],
  #     news: [True],
  #     date_time_stamp: [False],
  #     appointments: [False]
  #   }
  # ]