experiments/gnn_architecture_comparison.yaml

# Fleet Spec: Controlled GNN Architecture Comparison
#
# Direct comparison of 5 GNN architectures (GCN, SAGE, GAT, GIN, GraphConv)
# with controlled hyperparameters for Ruby code complexity prediction.
#
# Launch:
#   ratiocinator fleet run experiments/gnn_architecture_comparison.yaml
#
# Each arm varies ONLY the conv_type. All other hyperparameters are identical.

name: gnn-architecture-comparison
description: "5-way GNN architecture comparison on Ruby AST complexity prediction"

hardware:
  gpu: "RTX 4090"
  num_gpus: 1
  min_cpu_ram_gb: 32
  min_inet_down: 1000.0
  min_cuda_version: 12.0
  max_dph: 0.40
  disk_gb: 50.0
  image: pytorch/pytorch:2.7.0-cuda12.8-cudnn9-runtime

repo:
  url: https://github.com/timlawrenz/jubilant-palm-tree.git
  branch: experiment/ratiocinator-gnn-study
  clone_depth: 1

data:
  source: none

deps:
  pre_install:
    - "apt-get update -qq && apt-get install -y -qq git-lfs > /dev/null 2>&1 || true"
    - "cd /workspace/experiment && git lfs install && git lfs pull"
    - "pip install torch-geometric torch-scatter torch-sparse -f https://data.pyg.org/whl/torch-2.7.0+cu128.html"
    - "pip install pandas tqdm sentence-transformers nltk scikit-learn numpy"
  requirements: requirements.txt
  exclude_from_requirements:
    - torch
    - torchvision
    - torch_geometric
  verify: "python -c \"import torch_geometric; print(f'PyG {torch_geometric.__version__}')\""

arms:
  # ── Architecture comparison (same hyperparams, different conv) ──
  - name: sage-baseline
    description: "GraphSAGE baseline (original architecture)"
    command: "bash scripts/run_complexity_arm.sh"
    env:
      CONV_TYPE: "SAGE"
      HIDDEN_DIM: "64"
      NUM_LAYERS: "3"
      DROPOUT: "0.1"
      LEARNING_RATE: "0.001"
      EPOCHS: "50"

  - name: gcn
    description: "Graph Convolutional Network"
    command: "bash scripts/run_complexity_arm.sh"
    env:
      CONV_TYPE: "GCN"
      HIDDEN_DIM: "64"
      NUM_LAYERS: "3"
      DROPOUT: "0.1"
      LEARNING_RATE: "0.001"
      EPOCHS: "50"

  - name: gat
    description: "Graph Attention Network"
    command: "bash scripts/run_complexity_arm.sh"
    env:
      CONV_TYPE: "GAT"
      HIDDEN_DIM: "64"
      NUM_LAYERS: "3"
      DROPOUT: "0.1"
      LEARNING_RATE: "0.001"
      EPOCHS: "50"

  - name: gin
    description: "Graph Isomorphism Network"
    command: "bash scripts/run_complexity_arm.sh"
    env:
      CONV_TYPE: "GIN"
      HIDDEN_DIM: "64"
      NUM_LAYERS: "3"
      DROPOUT: "0.1"
      LEARNING_RATE: "0.001"
      EPOCHS: "50"

  - name: graphconv
    description: "GraphConv (Morris et al.)"
    command: "bash scripts/run_complexity_arm.sh"
    env:
      CONV_TYPE: "GraphConv"
      HIDDEN_DIM: "64"
      NUM_LAYERS: "3"
      DROPOUT: "0.1"
      LEARNING_RATE: "0.001"
      EPOCHS: "50"

  # ── Hyperparameter variants on best-expected architectures ──
  - name: sage-wide
    description: "SAGE with 128 hidden dim"
    command: "bash scripts/run_complexity_arm.sh"
    env:
      CONV_TYPE: "SAGE"
      HIDDEN_DIM: "128"
      NUM_LAYERS: "3"
      DROPOUT: "0.1"
      LEARNING_RATE: "0.001"
      EPOCHS: "50"

  - name: gat-wide
    description: "GAT with 128 hidden dim"
    command: "bash scripts/run_complexity_arm.sh"
    env:
      CONV_TYPE: "GAT"
      HIDDEN_DIM: "128"
      NUM_LAYERS: "3"
      DROPOUT: "0.1"
      LEARNING_RATE: "0.001"
      EPOCHS: "50"

  - name: sage-deep
    description: "SAGE with 5 layers"
    command: "bash scripts/run_complexity_arm.sh"
    env:
      CONV_TYPE: "SAGE"
      HIDDEN_DIM: "64"
      NUM_LAYERS: "5"
      DROPOUT: "0.1"
      LEARNING_RATE: "0.001"
      EPOCHS: "50"

metrics:
  protocol: json_line
  json_prefix: "METRICS:"

budget:
  max_dollars: 10.00
  train_timeout_s: 1200
  download_timeout_s: 600