DFA + residual-branch penalty test: seed=456, lam=0.01
  ep 0: ||h_L||=8.890e+00 ||g_2||=9.971e-04 acc=0.0928
  ep 1: ||h_L||=8.559e+02 ||g_2||=1.043e-05 eval_acc=0.3379 test_acc=0.3428
  ep 10: ||h_L||=8.393e+03 ||g_2||=2.127e-06 eval_acc=0.3564 test_acc=0.3525
  ep 20: ||h_L||=1.638e+04 ||g_2||=1.497e-06 eval_acc=0.3604 test_acc=0.3563
  ep 30: ||h_L||=2.349e+04 ||g_2||=1.225e-06 eval_acc=0.3662 test_acc=0.3606
  ep 40: ||h_L||=2.934e+04 ||g_2||=1.088e-06 eval_acc=0.3428 test_acc=0.3427
  ep 50: ||h_L||=3.396e+04 ||g_2||=1.004e-06 eval_acc=0.3447 test_acc=0.3538
  ep 60: ||h_L||=3.729e+04 ||g_2||=9.557e-07 eval_acc=0.3701 test_acc=0.3654
  ep 70: ||h_L||=3.938e+04 ||g_2||=9.202e-07 eval_acc=0.3760 test_acc=0.3655
  ep 80: ||h_L||=4.053e+04 ||g_2||=9.116e-07 eval_acc=0.3691 test_acc=0.3615
  ep 90: ||h_L||=4.096e+04 ||g_2||=8.987e-07 eval_acc=0.3682 test_acc=0.3639
  ep 100: ||h_L||=4.102e+04 ||g_2||=9.000e-07 eval_acc=0.3691 test_acc=0.3635

FINAL test acc: 0.3635
Compare to:
  DFA-vanilla (3-seed mean):    0.308
  DFA-shallow (3-seed mean):    0.349
  DFA-frozen (3-seed mean):     0.349
  BP-trainable (3-seed mean):   0.609
Saved results/dfa_residual_penalty/dfa_pen_lam0.01_s456.json