fnbm-order3-stacking-parity10-higher-order-v1

10-epoch performance check of the feature/higher-order chain-loop model on the 20260915 ordered-triplet no-noise simulation. Arms: control (orders 1+2) and o3_mask (orders 1+2+3, indirect self-interaction mask on). Seed 0, all orders active from epoch 0, dropout 0. One row per (arm, epoch); test metrics repeated per row.

Dataset Info

• Rows: 20
• Columns: 15

Columns

Column	Type	Description
arm	Value('string')	control_s0 (orders 1+2) or o3_mask_s0 (orders 1+2+3, indirect mask on)
epoch	Value('int64')	0-indexed training epoch
train_loss	Value('float64')	training loss incl. penalties
val_loss	Value('float64')	validation MSE
val_pearson_r	Value('float64')	validation Pearson r
order1_contribution_l2	Value('float64')	RMS of order-1 output (last training batch)
order2_contribution_l2	Value('float64')	RMS of order-2 per-position output
order3_contribution_l2	Value('float64')	RMS of order-3 per-position output (NaN for control)
test_loss	Value('float64')	test MSE of best checkpoint
test_pearson_r	Value('float64')	test Pearson r of best checkpoint
best_epoch	Value('int64')	epoch of best val loss
job_id	Value('string')	slurm array task
config_yaml	Value('string')	full resolved training config
sbatch_script	Value('string')	sbatch used
git_commit	Value('string')	feature/higher-order commit run

Generation Parameters

{
  "script_name": "train.py",
  "model": "FactorizedNBM (feature/higher-order order-n chain, commit 5b78f54)",
  "description": "10-epoch performance check of the feature/higher-order chain-loop model on the 20260915 ordered-triplet no-noise simulation. Arms: control (orders 1+2) and o3_mask (orders 1+2+3, indirect self-interaction mask on). Seed 0, all orders active from epoch 0, dropout 0. One row per (arm, epoch); test metrics repeated per row.",
  "hyperparameters": {
    "num_epochs": 10,
    "seed": 0,
    "order_start_epoch": "all 0",
    "mask_indirect_self_interactions": true,
    "lr": 0.001,
    "batch_size": 128
  },
  "input_datasets": [
    "20260915_triplet_no_noise (torch scratch)"
  ],
  "experiment_name": "fnbm-order3-stacking",
  "job_id": "torch:18601925",
  "cluster": "torch",
  "artifact_status": "final",
  "canary": false
}

Usage

from datasets import load_dataset

dataset = load_dataset("arushram/fnbm-order3-stacking-parity10-higher-order-v1", split="train")
print(f"Loaded {len(dataset)} rows")

---