Data:
  # Basics
  log_dir: 'tasks/models'
  # Data
  dataset: "FFTDataset"
  data_dir: None
  model_name: "CNNEncoder"
  batch_size: 4
  num_epochs: 10
  exp_num: 2
  max_len_spectra: 4096
  max_days_lc: 270
  lc_freq: 0.0208
  create_umap: True
  checkpoint_path: 'tasks/models/frugal_2025-01-29/frugal_kan_2.pth'

CNNEncoder:
  # Model
  in_channels: 2
  num_layers: 4
  stride: 1
  encoder_dims: [32,64,128]
  kernel_size: 3
  dropout_p: 0.3
  output_dim: 2
  beta: 1
  load_checkpoint: False
  checkpoint_num: 1
  activation: "silu"
  sine_w0: 30.0
  avg_output: False

MLP:
  input_dim: 6
  hidden_dims: [16,32,6]
  dropout: 0.2

KAN:
  layers_hidden: [1125,32,8,1]
  grid_min:  -1.2
  grid_max:  1.2
  num_grids:  8
  exponent:  2

KAN_INR:
  layers_hidden: [1,1024,128,128,1]
  grid_min:  -1.2
  grid_max:  1.2
  num_grids:  8
  exponent:  2

CNNEncoder_f:
  # Model
  in_channels: 32
  num_layers: 4
  stride: 1
  encoder_dims: [32,64,128]
  kernel_size: 3
  dropout_p: 0.3
  output_dim: 2
  beta: 1
  load_checkpoint: True
  checkpoint_num: 1
  activation: "silu"
  sine_w0: 1.0
  avg_output: True


Conformer:
  encoder: ["mhsa_pro", "conv"]
  timeshift: false
  num_layers: 4
  encoder_dim: 128
  num_heads: 8
  kernel_size: 3
  dropout_p: 0.2
  norm: "postnorm"

RelationalTransformer:
  d_node: 32
  d_edge: 32
  d_attn_hid: 16
  d_node_hid: 16
  d_edge_hid: 16
  d_out_hid: 16
  d_out: 1
  n_layers: 4
  n_heads: 4
  dropout: 0.1


INR:
  in_features : 2
  n_layers : 2
  hidden_features : 64
  out_features : 32

XGBoost:
  objective : 'binary:logistic'
  eval_metric : 'logloss'
  use_label_encoder : False
  n_estimators : 500
  learning_rate : 0.1
  max_depth : 5
  subsample : 0.8
  colsample_bytree : 0.8
  random_state : 42

Optimization:
  # Optimization
  max_lr: 1e-5
  weight_decay: 5e-6
  warmup_pct: 0.3
  steps_per_epoch: 3500