bambezius/dissertation-gnn-sym

gnn-sym

JAX checkpoint for the GatV2Sym D2-invariant shared-trunk graph model with separate intention and outcome heads.

Checkpoint

• Model class: GatV2Sym
• Framework: jax
• Run timestamp: 20260430_113740
• Parameters: model_params.joblib
• Hyperparameters: config.json
• Metrics: metrics.json

Validation Metrics

{
  "val/loss": 1.0811628103256226,
  "val/intention_top1_accuracy": 0.6713362336158752,
  "val/intention_top2_accuracy": 0.8514277935028076,
  "val/intention_top3_accuracy": 0.9220097064971924,
  "val/intention_mrr": 0.8004188537597656,
  "val/outcome_top1_accuracy": 0.6073545217514038,
  "val/outcome_top2_accuracy": 0.7707435488700867,
  "val/outcome_top3_accuracy": 0.8465786576271057,
  "val/outcome_mrr": 0.7399893403053284
}

Loading

import json
from pathlib import Path

import joblib

checkpoint_path = Path("path/to/checkpoint")

with open(checkpoint_path / "config.json") as f:
    config = json.load(f)

params = joblib.load(checkpoint_path / "model_params.joblib")