bambezius/dissertation-tgn-sym-dynamic

tgn-sym-dynamic

JAX checkpoint for the TgnSymDynamic football pass prediction model.

Checkpoint

• Model class: TgnSymDynamic
• Framework: jax
• Run timestamp: 20260430_084243
• Parameters: model_params.joblib
• Hyperparameters: config.json
• Metrics: metrics.json

Validation Metrics

{
  "val/loss": 0.928921639919281,
  "val/intention_top1_accuracy": 0.7339416146278381,
  "val/intention_top2_accuracy": 0.8846715688705444,
  "val/intention_top3_accuracy": 0.9397810697555542,
  "val/intention_mrr": 0.8394966125488281,
  "val/outcome_top1_accuracy": 0.6613138914108276,
  "val/outcome_top2_accuracy": 0.8071168065071106,
  "val/outcome_top3_accuracy": 0.8708029389381409,
  "val/outcome_mrr": 0.777962327003479
}

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")