# Research Spec: GNN Architecture Search for Code Complexity Prediction
#
# Automated search across 5 GNN architectures and hyperparameters
# for predicting Ruby method complexity from AST structure.
#
# Launch:
#   ratiocinator research specs/gnn_complexity.yaml
#
# The dataset (22K Ruby methods) is in the repo branch.
# No external data staging needed.

# What to research
topic: "Comparing GNN architectures (GCN, SAGE, GAT, GIN, GraphConv) for predicting Ruby code complexity from Abstract Syntax Trees. The baseline uses GraphSAGE with hidden_dim=64, 3 layers, achieving MAE 4.77 on 22K Ruby methods."
goal_metric: val_mae
maximize: false

# Target codebase
repo_url: https://github.com/timlawrenz/jubilant-palm-tree.git
repo_branch: experiment/ratiocinator-gnn-study
runner_script: scripts/run_complexity_arm.sh

# Infrastructure — models are tiny (~50K params), training is fast
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

data:
  source: none  # Dataset is in the repo branch

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__}')\""

metrics:
  protocol: json_line
  json_prefix: "METRICS:"

# Budget — small models, fast training (~2-5 min per arm)
max_iterations: 3
max_dollars: 15.00
train_timeout_s: 1200
download_timeout_s: 600

# Output
paper_title: "What Graph Neural Networks Can and Cannot Learn About Code: A Systematic Empirical Study on Ruby AST Analysis"