config_hyperopt.yml

# config_hyperopt.yml
# Configuration for a full hyperparameter optimization search using Hyperopt.

# Global settings for the hyperopt run
global_params:
  verbose: 1 # Less verbose for multiple runs
  error_raise: true # Don't stop the search on a single trial failure
  # Set to true to see H2O's progress bars for parsing and model fitting.
  h2o_show_progress: false
  # Number of iterations for RandomizedSearchCV and BayesSearchCV
  n_iter: 2
  max_param_space_iter_value : 10
  sub_sample_param_space_pct: 0.05
  force_second_cv: false # If True, forces a second cross-validation run even if cached results are available. Defaults to False.
  model_eval_time_limit: 3600
  test_mode: true # If True, uses minimal parameter spaces and reduced cross-validation for fast testing.

# Experiment settings for the hyperopt run
experiment:
  # Base directory for storing all experiment results
  experiments_base_dir: "HFE_ML_experiments"
  # A descriptive name for this experiment batch
  additional_naming: ""

# Data settings for the hyperopt run
data:
  # Path to the dataset
  file_path: "synthetic_data_generated.csv"
  # List of substrings to identify columns to drop
  drop_term_list:
    - 'chrom'
    - 'hfe'
    - 'phlebo'
  # Set to true to iterate over all 'outcome_var_' columns in the dataset
  multiple_outcomes: true

# Models to be used during the hyperopt search
models:
  LogisticRegressionClass: true
  KNeighborsClassifierClass: true
  QuadraticDiscriminantAnalysisClass: true
  SVCClass: true
  XGBClassifierClass: true
  MLPClassifierClass: true
  RandomForestClassifierClass: true
  GradientBoostingClassifierClass: true
  CatBoostClassifierClass: true
  GaussianNBClassifierClass: true
  H2O_class: false # H2O AutoML
  H2O_GBM_class: true
  H2O_DRF_class: true
  H2O_DeepLearning_class: true
  H2O_GLM_class: true
  H2O_NaiveBayes_class: true
  H2O_RuleFit_class: true
  H2O_XGBoost_class: true
  H2O_StackedEnsemble_class: false
  H2O_GAM_class: true
  LightGBMClassifierWrapper: true # noqa
  AdaBoostClassifierClass: true
  NeuralNetworkClassifier_class: true # noqa
  TabTransformerClass: false
  # Set the following to true if a GPU is available and configured
  kerasClassifier_class: false
  knn__gpu_wrapper_class: false
  # Additional
  TabPFNClassifierClass: false # requires hf token and agreement
  AutoGluonClassifierClass: false # disabled by default
  TPOTClassifierClass: false # disabled by default
  FLAMLClassifierClass: false # disabled by default
  AutoKerasClassifierClass: false # disabled by default


# Time-series models to be used during the hyperopt search.
# These are only loaded if time_series_mode is enabled in the pipeline.
ts_models:
  # Enabled by default
  RocketClassifier: true
  TimeSeriesForestClassifier: true
  KNeighborsTimeSeriesClassifier: true
  Catch22Classifier: true
  HIVECOTEV2: false # Disabled due to internal numba typing error in aeon library
  TSFreshClassifier: true
  Arsenal: true
  CNNClassifier: true
  ElasticEnsemble: true
  EncoderClassifier: true
  FCNClassifier: true
  FreshPRINCEClassifier: true
  InceptionTimeClassifier: false # Disabled due to Keras input shape mismatch error
  IndividualInceptionClassifier: false # Disabled due to Keras input shape mismatch error
  IndividualTDE: true
  MLPClassifier: true
  MUSE: true
  OrdinalTDE: true
  ResNetClassifier: true
  SignatureClassifier: true
  SummaryClassifier: true
  TemporalDictionaryEnsemble: true
  # Univariate models, disabled by default
  ContractableBOSS: false
  HIVECOTEV1: false

# This section defines the parameter search space for Hyperopt.
# The structure uses lists of options, which will be parsed into hp.choice.
hyperopt_search_space:
  resample: ['undersample', 'oversample', 'null']
  scale: [true, false]
  feature_n: [100, 95, 75, 50, 25, 5]
  param_space_size: ['medium', 'xsmall']
  percent_missing: [99, 95, 80]
  correlation_threshold: [0.98, 0.85, 0.5, 0.25]
  feature_selection_method: ['anova', 'markov_blanket']
  outcome_var_n: ['1'] # Default, will be overridden if multiple_outcomes is true

  # Embedding hyperparameters
  use_embedding: [true, false]
  embedding_method: ['pca', 'svd']
  embedding_dim: [32, 64, 128]
  scale_features_before_embedding: [true, false]

  # Feature group toggles
  data:
    age: [true, false]
    sex: [true, false]
    bmi: [true]
    ethnicity: [true, false]
    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: [False]
    bed: [False]
    vte_status: [true]
    hosp_site: [true]
    core_resus: [False]
    news: [False]
    date_time_stamp: [False]
    appointments: [False]

# Hyperopt-specific settings
hyperopt_settings:
  max_evals: 2 # Number of iterations per outcome variable
  trial_timeout: 1120 # Timeout in seconds for a full trial (data prep + all models)