params.yaml

experiment:
  EXPERIMENT_NAME: 'MLP_testing'  # Name of the experiment
  TRACKING_URI : "https://dagshub.com/sarthakg004/convolve.mlflow"
##################################################################################################################################################
data_cleaning:
  DEV_DATA_PATH : 'data/raw/Dev_data_to_be_shared.csv'  # Path to the raw data
  VAL_DATA_PATH : 'data/raw/validation_data_to_be_shared.csv'  # Path to the raw validation data

  ENCODING_TECHNIQUE : 'target'  # Encoding technique to be used for categorical features  'target' or 'ordinal'
  
  NULL_PERCENTAGE : 0.25   # features with more than given missing values will be dropped
  VARIANCE_THRESHOLD : 0   # features with less variance than given threshold will be dropped
  IMPUTATION_TECHNIQUE : 'mode'  # Impuation technique to be used for missing values
  KNN_IMPUTER_N_NEIGHBORS : 5    # Number of neighbors to be used for KNN imputation
  TEST_SIZE : 0.2     # Test size for train-test split
  RANDOM_STATE : 42
##################################################################################################################################################
feature_engineering:

  INTERIM_TRAIN_DATA_PATH : 'data/interim/train.csv'  # Path to the interim train data
  INTERIM_TEST_DATA_PATH : 'data/interim/test.csv'  # Path to the interim test data
  INTERIM_VAL_DATA_PATH : 'data/interim/validation.csv'  # Path to the interim validation data


  FEATURE_IMPORTANCE_TECHNIQUE : 'combine'  # Feature importance technique to be used 'combine' or 'fisher' or 'rf' or 'xgb' 
  NO_FEATURES : 40

  FEATURE_SELECTION_TECHNIQUE : 'feature_importance'  # Feature selection technique to be used    'anova' or 'backward' or 'feature_importance' or'skip'
  ANOVA_K : 40  # Number of features to be selected using ANOVA

  # If selection technique is Sequential Backward Selection than hyperparameters of RF model
  SFS_FORWARD : False  # IF False than SFS will be used as SBS
  N_ESTIMATORS : 200
  MAX_DEPTH : 10
  MIN_SAMPLES_SPLIT : 5
  MIN_SAMPLES_LEAF : 2
  MAX_FEATURES : 'sqrt'
  RANDOM_STATE : 42

############################################################################################################################################################

model_training:

  PROCESSED_TRAIN_DATA : 'data/processed/train.csv'  # Path to the processed train data
  PROCESSED_TEST_DATA : 'data/processed/test.csv'  # Path to the processed test data
  PROCESSED_VAL_DATA : 'data/processed/val.csv'  # Path to the processed validation data

  MODEL : 'mlp'  # Model to be used for training

  # Hyperparameters Space of XGB model

  N_TRIALS : 10
  
  MAX_DEPTH : [3,15]
  OBJECTIVE: 'binary:logistic'
  EVAL_METRIC: 'logloss'
  ETA  : [0.005, 0.3]
  SUB_SAMPLE : [0.5, 1.0]
  COL_SAMPLE_BY_TREE : [0.5, 1.0]
  MIN_CHILD_WEIGHT: [1, 20]
  GAMMA_XGB :  [0, 5] # Gamma range
  LAMBDA : [0, 10]  # L2 regularization range
  ALPHA : [0, 10] # L1 regularization range
  SCALE_POS_WEIGHT : [1, 10]   # Class imbalance correction
  N_ESTIMATORS: [50, 500]  # Number of boosting rounds
  TREE_METHOD: 'hist'
  DEVICE: 'cuda'
  SEED: 42

  

  # Hyperparameters of MLP model
  TRIALS : 1
  NO_LAYERS : [2,10]
  HIDDEN_DIMS : [32,512,32] # Hidden dimensions of each layer [ start, end , step]
  DROPOUT_RATE : [0.1,0.5,0.1] # Dropout rate of each layer [ start, end , step]
  LEARNING_RATE : [0.00001,0.1] # Learning rate range
  WEIGHT_DECAY : [0.00001, 0.001]
  BATCH_SIZE : [16,32,64,128,256]  # List of batch sizes
  N_EPOCHS : [10,50,10]  # Number of epochs [ start, end , step]
  OPTIMIZER : ["Adam", "SGD", "RMSprop"]

##################################################################################################################################
prediction:
  DATA : 'data/processed/val.csv'  # Path to the processed test data
  MODEL : 'mlp'  # Model to be used for prediction
  XGB_MODEL_PATH : 'models/xgboost_model.pkl'  # Path to the trained XGB model
  MLP_MODEL_PATH : 'models/mlp_model.pth'  # Path to the trained MLP model
  SUBMISSION_FILE_PATH : 'prediction.csv'  # Path to the submission file