DAMA-BAX/config.example.yaml at main · SLAC-ML/DAMA-BAX · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
# DAMA-BAX Configuration Example
# Copy this file to config.yaml and modify as needed

# Resource paths
resources:
  base_dir: "./resources"
  matlab_data: "${resources.base_dir}/matlab_data"
  setup: "${resources.base_dir}/setup"
  rings: "${resources.base_dir}/rings"
  pretrained_models: "${resources.base_dir}/models/run_0"
  initial_data: "${resources.base_dir}/data/run_0"

# Run configuration
run:
  name: "run_3"  # Identifier for this optimization run
  data_root: "./data/${run.name}/"
  model_root: "./models/${run.name}/"
  log_filename: "bax_log_${run.name}.pkl"
  max_iterations: 3200  # Maximum number of BAX loops
  n_sampling: 50  # Number of points sampled per iteration

# Problem parameters
problem:
  da:
    threshold: 0.75  # DA threshold (survival fraction defining boundary)
    method: 1  # Calculation method: 0=Daniel's, 1=Xiaobiao's
  ma:
    threshold: 0.94  # MA threshold
    method: 2  # Calculation method: 0=Daniel's, 1=Xiaobiao's, 2=Xiaobiao's (GT match)

# Genetic algorithm parameters
ga:
  population_size: 200  # NSGA2 population size
  n_generations: 20  # Number of generations
  selection_size: 100  # Number of solutions selected after optimization

# Acquisition strategy
acquisition:
  method: 2  # 0=around boundary, 1=at boundary, 2=within survival range
  da_range:  # For method=2: survival turn range for DA
    lower_bound: 0.4
    upper_bound: 0.75
  ma_range:  # For method=2: survival turn range for MA
    lower_bound: 0.85
    upper_bound: 0.95

# Neural network training parameters
training:
  # Initial training (on pretrained models)
  initial:
    epochs: 150
    learning_rate: 0.0001
    batch_size: 1000
    dropout: 0.1
    n_neurons: 800
    early_stopping_patience: 10

  # Iterative training (during BAX loop)
  iterative:
    epochs: 10
    learning_rate: 0.0001
    batch_size: 1000
    weight_new_points: 10  # Weight multiplier for newly acquired points

  # Training settings
  test_ratio: 0.05
  random_state: 1

# Computation settings
compute:
  device: "auto"  # "auto", "cuda", "cpu"
  max_inference_points: 100000  # Maximum points per inference batch
  simulation_buffer_size: 200  # Buffer size for simulation data saving
  num_workers: 1  # DataLoader workers

# Tracking simulation parameters
simulation:
  n_turns: 1024  # Number of tracking turns (512 * 2)
  seeds: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]  # Random seeds for stochastic simulations

# Checkpointing
checkpointing:
  enabled: true
  save_snapshots: true  # Save model at every iteration (vs. only best)
  save_final: true  # Save final model state

# Logging
logging:
  verbose: 1  # 0=quiet, 1=normal, 2=debug
  log_period: 10  # Print every N batches during training