save_name: "xgboost_OS"
target: "OS"


#### Data paths + pre-processing for each modality (e.g. simple imputation)
clinical_data:
  clinical_file: "data/clinicals.csv"
  imputation:
    numericals: [1, 3, 4, 13, 14, 15, 16, 17, 18, 19, -4]
    categoricals: [20, -5, -3]

radiomics_data:
  radiomics_file: "data/radiomics.csv"
  preprocessing:
    f_log_transform: ['TMTV', 'T_TMTV', 'N1_TMTV', 'N2_TMTV', 'N3_TMTV', 'M1a_TMTV', 'M1bc_TMTV']
  imputation:
    numericals: [2]
    categoricals: null

pathomics_data:
  pathomics_file: "data/pathomics.csv"
  imputation:
    numericals: [ 27, 28, 29, 31, 33, 35, 38, 40, 41, 43, 44, 45, 47, 48, 49, 52, 53, 54, 55, 56, 57, 58, 59, 60,
                  61, 62, 63, 64, 65, 66, 67, 118, 119, 120, 121, 122, 123, 124, 125, 126, 127, 128, 129, 130,
                  131, 132, 133 ]
    categoricals: null

RNA_data:
  RNA_file: "data/omics.csv"
  imputation:
    numericals: [32]
    categoricals: null


####  Define classifier + hyperparameters
classifier:
#  type: "LR"
#  args:
#    penalty: "elasticnet"
#    max_iter: 2500
#    solver: "saga"
#    class_weight: "balanced"
#    C: 0.1
#    l1_ratio: 0.5
  type: "xgboost"
  args:
    n_jobs: 1

  optim_params: # Uncomment and modify for parameter optimization
#     LR__l1_ratio: [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9]
#     LR__C: [0.001, 0.005, 0.01, 0.05, 0.1, 0.5, 1, 5, 10]
#
#     xgboost__max_depth: [3, 6, 9, 12]
#     xgboost__learning_rate: [0.001, 0.01, 0.05, 0.1, 0.2, 0.3]
#     xgboost__subsample: [0.5, 0.6, 0.7, 0.8, 0.9, 1.]
#     xgboost__colsample_bytree: [0.5, 0.6, 0.7, 0.8, 0.9, 1.]
#     xgboost__n_estimators: [100, 200, 300, 400, 500]
#  n_iter_randomcv: 80 # number of iterations/samples for randomsearch


#### Define latefusion parameters  (see multipit.multi_model.latefusion)
latefusion:
  n_repeats: 100 # number of repeats for 10-fold cross-validation
  seed: 3 # random seed
  args:
    sup_weights: False # learn weights with inner cross-validation for weighted late fusion
    calibration: 'late' # if 'late' calibrate the fused model, if 'early' calibrate each unimodal model before fusion
    n_jobs: 1
    # tuning: "gridsearch" # uncomment for parameter optimization
    # score: "roc_auc" # uncomment for parameter optimization


##### Additionnal parameters
parallelization: # number of jobs for dealing with several repeats in parallel (conflict with n_jobs of latefusion)
  n_jobs_repeats: 1

collect_thresholds: False  # collect threshold that optimizes log-rank test on the training set

permutation_test: False # perform permutation test
n_permutations: 1