ANALYSIS:
  DRAW_NEURON_LOCATION: false
  SAVE_NEURON_LOCATION: false
CHECKPOINT_DIR: /data/ckpt/
DATAMODULE:
  BATCH_SIZE: 2
  FEATURE_EXTRACTOR_MODE: false
  NUM_WORKERS: 8
  PIN_MEMORY: true
DATASET:
  CACHE_DIR: /data/cache
  DARK_POSTFIX: 
  FILTER_BY_SESSION:
  - -1
  FMRI_SPACE: full_fsaverage
  IMAGE_RESOLUTION:
  - 224
  - 224
  N_PREV_FRAMES: 32
  ROIS:
  - all
  ROOT: /data/ALG23
  SUBJECT_LIST:
  - subj01
  - subj02
  - subj03
  - subj04
  - subj05
  - subj06
  - subj07
  - subj08
DESCRIPTION: for alex
EXPERIMENTAL:
  ANOTHER_SPLIT: false
  BEHV_ONLY: false
  BEHV_SELECTION:
  - -1
  BLANK_IMAGE: false
  NO_SPLIT: false
  SHUFFLE_IMAGES: false
  SHUFFLE_VAL: true
  STRAIGHT_FORWARD: false
  STRAIGHT_FORWARD_BUT_KEEP_BACKBONE_GRAD: false
  T_IMAGE: 0
  USE_DEV_MODEL: true
  USE_PREV_FRAME: false
  USE_PREV_BEHV: false
  USE_BHV: false
  USE_BHV_PASSTHROUGH: false
  USE_CURRENT_FRAME: true
  USE_CURRENT_BEHV: false
  USE_FTR_FRAME: false
  USE_FTR_BEHV: false
LOSS:
  DARK:
    MAX_EPOCH: 150
    USE: false
  NAME: SmoothL1Loss
  SMOOTH_L1_BETA: 0.01
  SYNC:
    EMA_BETA: 0.9
    EMA_BIAS_CORRECTION: false
    EMA_KEY: running_grad
    EXP_SCALE: 10.0
    EXP_SHIFT: 0.0
    LOG_SHIFT: 10.0
    SKIP_EPOCHS: 20
    STAGE: VAL
    UPDATE_RULE: raw
    USE: false
MODEL:
  BACKBONE:
    ADAPTIVE_LN:
      SCALE: 0.5
    CACHE_DIR: /data/cache
    CLS_DIMS:
    - 1536
    - 1536
    - 1536
    - 768
    FEATURE_DIMS:
    - 768
    - 768
    - 768
    - 768
    LAYERS:
    - 2
    - 5
    - 8
    - 11
    LORA:
      RANK: 4
      SCALE: 0.2
    NAME: dinov2_vit_b
  BACKBONE_SMALL:
    ADAPTIVE_LN:
      SCALE: 0.5
    CLS_DIMS:
    - 384
    LAYERS:
    - 11
    LORA:
      RANK: 4
      SCALE: 0.2
    MERGE_WIDTH: 64
    NAME: dinov2_vit_s
    T_DIM: 64
    WIDTH: 64
  BOTTLENECK:
    OUT_DIM: 64
    RANK: -1
  COND:
    DIM: 256
    DROPOUT: 0.2
    IN_DIM: 35
    PASSTHROUGH_DIM: 32
    USE: true
  CONV_HEAD:
    DEPTHS:
    - 3
    - 3
    - 3
    - 3
    KERNEL_SIZES:
    - 5
    - 5
    - 5
    - 5
    MAX_DIM: 1024
    SIMPLE: true
    WIDTH: 192
  COORDS_MLP:
    DEPTH: 3
    LOG: false
    WIDTH: 128
  LAYER_SELECTOR: {}
  MAX_TRAIN_VOXELS: 25600
  PREV_FEAT:
    DIM: 1024
  RETINA_MAPPER:
    CONSTANT_SIGMA: 0.01
  VOXEL_OUTS:
    SHARED:
      MLP:
        DEPTH: 3
        WIDTH: 1024
      USE: false
MODEL_SOUP:
  GREEDY_TARGET: heldout
  RECIPE: greedy
  USE: true
OPTIMIZER:
  LR: 0.0001
  NAME: AdamW
  SCHEDULER:
    CYCLE_DECAY: 0.5
    CYCLE_LIMIT: 3
    K_DECAY: 1.5
    LR_MIN: 0.0001
    LR_MIN_WARMUP: 0.0001
    T_INITIAL: 1
    T_MULT: 1.0
    WARMUP_T: 10
  WEIGHT_DECAY: 0.0003
POSITION_ENCODING:
  FEATURES: 32
  IN_DIM: 3
  MAX_STEPS: 1000
  PERIODS: 10000
REGULARIZER:
  LAYER: 3.0e-05
RESULTS_DIR: /nfscc/alg23/ray_results/
TRAINER:
  ACCUMULATE_GRAD_BATCHES: 16
  CALLBACKS:
    CHECKPOINT:
      LOAD_BEST_ON_END: false
      LOAD_BEST_ON_VAL: false
      REMOVE: true
      SAVE_TOP_K: 10
    EARLY_STOP:
      PATIENCE: 60
  DDP: false
  GRADIENT_CLIP_VAL: 0.5
  LIMIT_TRAIN_BATCHES: 0.1
  LIMIT_VAL_BATCHES: 0.5
  LOG_TRAIN_N_STEPS: 100
  MAX_EPOCHS: 500
  MAX_STEPS: -1
  PRECISION: 16
  VAL_CHECK_INTERVAL: 1.0