# ============================================================
# Task: lane_count_regression
# ============================================================
# Task objective:
# - Predict the lane count, Lanes, from static attributes of LargeST road segments / sensors.
# - This is an episode-level regression task: each ID corresponds to one static sample.
#
# Core semantics:
# - sample is a road_segment / sensor ID.
# - inputs are the static attributes available to the model.
# - prediction.target.column = Lanes is the regression target the model should predict.
# - Lanes must not be an input field, as this would directly leak the label.
# ============================================================

task:
  id: lane_count_regression # Unique task ID; keep it consistent with the directory name.
  type: episode_regression # Task type: output one continuous value or numeric label per sample.
  description: >
    Predict the number of lanes for each LargeST road segment / sensor episode
    using static metadata features.

inherit_general: true

sample:
  primary_key: ID # ID joins labels.csv, splits.csv, and the static attribute table.

inputs:
  # ------------------------------------------------------------
  # Static tabular inputs.
  # ------------------------------------------------------------
  tabular_static:
    source: largeST_static_metadata # Refers to the static attribute table in data_refs.tabular_static.

    # Explicitly list the static fields available to the model.
    # The following fields are common in LargeST / PeMS road metadata and may help predict lane count.
    # If the column names in your metadata file differ, update them here accordingly.
    fields:
      - Lat
      - Lng
      - District
      - County
      - Fwy
      - Lanes
      - Type
      - Direction


  time_series:
    - fields: traffic_flow  # Historical time series fields available to the model.
      source: traffic_flow_ca_2019 # Refers to the underlying LargeST h5 file defined in data_refs.time_series.

      window:
        length: 1344
        unit: 15min

  graph:
      source: road_distance_graph


prediction:
  field: Lanes # Supervision column in labels.csv.


leakage:
  forbidden_input_columns:
    - Lanes # Target column; it must never be used as an input.
    - ID # The primary key is used only for joins / indexing.

  # If Fwy reveals too much about lane count, move it here from inputs.tabular_static.fields.
  #   - Fwy

data_refs:
  tabular_static:
    largeST_static_metadata:
      file: All_Datasets/LargeST_ca_full/input_multimodal/ca_meta.csv # Adjust this to the actual file location; relative paths are preferred.
      id_column: ID

  time_series:
    traffic_flow_ca_2019:
      file: All_Datasets/LargeST_ca_full/input_multimodal/time_series/traffic_flow/ca_his_2019.h5
      value_field: traffic_flow
      native_grid_minutes: 15

  graph:
    road_distance_graph:
      file: All_Datasets/LargeST_ca_full/input_multimodal/node_level_task_graphs/road_distance_graph/edge_infos.csv