DerivSAC/docker-compose.yml at main · Thenolypus/DerivSAC · 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
services:
  # Database initialization container (simplified)
  db-init:
    build: .
    volumes:
      - ./logs:/app/logs
      - ./trading_bot/data/local_data:/app/trading_bot/data/local_data
    command: python -c "import os; import optuna; os.makedirs('logs/optuna_storage', exist_ok=True); optuna.create_study(study_name='nvidia_trading_optimization', direction='maximize', storage='sqlite:///logs/optuna_storage/optimization.db', load_if_exists=True); print('Study created successfully')"

  # Optuna Dashboard
  optuna-dashboard:
    image: ghcr.io/optuna/optuna-dashboard
    ports:
      - "8080:8080"
    volumes:
      - ./logs/optuna_storage:/app
    working_dir: /app
    depends_on:
      - db-init
    command: sqlite:///optimization.db

  trading-bot:
    build: .
    volumes:
      - ./logs:/app/logs
      - ./trading_bot/data/local_data:/app/trading_bot/data/local_data
    # GPU support with optimized settings
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]
    restart: unless-stopped
    depends_on:
      - db-init
    environment:
      # CPU Optimization for Ryzen 9 5950x (16 cores, 32 threads)
      - CUDA_VISIBLE_DEVICES=0
      - OMP_NUM_THREADS=32              # Utilize all 32 threads
      - MKL_NUM_THREADS=32              # Intel MKL threading
      - NUMBA_NUM_THREADS=32            # Numba threading
      - OPENBLAS_NUM_THREADS=32         # OpenBLAS threading
      - VECLIB_MAXIMUM_THREADS=32       # vecLib threading

      # GPU Optimization for RTX 5060 TI
      - PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:512,garbage_collection_threshold:0.6
      - CUDA_LAUNCH_BLOCKING=0          # Asynchronous CUDA operations
      - TF_FORCE_GPU_ALLOW_GROWTH=true  # Dynamic GPU memory allocation
      - TF_GPU_ALLOCATOR=cuda_malloc_async

      # Python Optimization
      - PYTHONUNBUFFERED=1              # Better logging output
      - PYTHONDONTWRITEBYTECODE=1       # Reduce I/O overhead
      - NVIDIA_OPTIMIZATION_MODE=true   # Enable optimization mode

      # Memory Management
      - MALLOC_ARENA_MAX=4              # Limit memory arenas for better performance
    command: ["python", "-m", "trading_bot.sac.zoo.train",
              "--algo", "sac",
              "--env", "NVIDIAKnockout-v0",
              "--device", "cuda",
              "--optimize",

              # Optimization Parameters - Scaled for 16-core CPU
              "--n-trials", "300",                    # Increased trials
              "--max-total-trials", "300",
              "--n-jobs", "20",                       # Increased parallel jobs (16 cores + buffer)
              "--sampler", "tpe",
              "--pruner", "median",

              # Evaluation Parameters - More thorough evaluation
              "--n-evaluations", "6",                 # Increased evaluations per trial
              "--eval-episodes", "12",                # More evaluation episodes
              "--eval-freq", "15000",                 # Slightly less frequent but more thorough

              # Configuration and Storage
              "--conf-file", "hyperparameters/sac.yml",
              "--storage", "sqlite:///logs/optuna_storage/optimization.db",
              "--study-name", "nvidia_trading_optimization",
              "--optimization-log-path", "logs/optimizer_results"]