future_window:
  value: 5
past_window:
  value: 22
batch_size:
  value: 64
max_epochs:
  value: 30
data:
  value:
    # in_sample_artifact_name: "btc-5m-features-in_sample:latest" # Reverted
    # out_of_sample_artifact_name: "btc-5m-features-out_of_sample:latest" # Reverted
    dataset: "btc-5m-features-full:latest" # Use a single artifact name
    sliding_window: 0
    validation: 0.2 # This likely controls the in-sample vs out-of-sample split in train.py
fields:
  value:
    time_index: "time_index"
    target: "returns"
    group_ids: ["group_id"]
    dynamic_unknown_real:
      - "high_price"
      - "low_price"
      - "open_price"
      - "close_price"
      - "volume"
      - "open_to_close_price"
      - "high_to_close_price"
      - "low_to_close_price"
      - "high_to_low_price"
      - "returns"
      - "log_returns"
      - "vol_1h"
      - "macd"
      - "macd_signal"
      - "rsi"
      - "low_bband_to_close_price"
      - "up_bband_to_close_price"
      - "mid_bband_to_close_price"
      - "sma_1h_to_close_price"
      - "sma_1d_to_close_price"
      - "sma_7d_to_close_price"
      - "ema_1h_to_close_price"
      - "ema_1d_to_close_price"
    dynamic_unknown_cat: []
    dynamic_known_real:
      - "effective_rates"
      - "vix_close_price"
      - "fear_greed_index"
      - "vol_1d"
      - "vol_7d"
    dynamic_known_cat:
      - "hour"
      - "weekday"
    static_real: []
    static_cat: []
loss:
  value:
   name: "Quantile"
   quantiles: [0.01, 0.02, 0.03, 0.05, 0.1, 0.25, 0.5, 0.75, 0.9, 0.95, 0.97, 0.98, 0.99]
model:
  value:
    name: "Informer"
    d_model: 256
    d_fully_connected: 512
    n_attention_heads: 2
    dropout: 0.05
    n_encoder_layers: 1
    n_decoder_layers: 1
    learning_rate: 0.0001
    optimizer: "Adam"