DFA + residual-branch penalty test: seed=42, lam=0.001
  ep 0: ||h_L||=8.893e+00 ||g_2||=9.934e-04 acc=0.1152
  ep 1: ||h_L||=1.232e+03 ||g_2||=7.279e-06 eval_acc=0.3232 test_acc=0.3225
  ep 10: ||h_L||=8.451e+03 ||g_2||=1.682e-06 eval_acc=0.3477 test_acc=0.3521
  ep 20: ||h_L||=1.571e+04 ||g_2||=1.124e-06 eval_acc=0.3809 test_acc=0.3648
  ep 30: ||h_L||=2.251e+04 ||g_2||=9.157e-07 eval_acc=0.3799 test_acc=0.3637
  ep 40: ||h_L||=2.826e+04 ||g_2||=8.129e-07 eval_acc=0.3906 test_acc=0.3630
  ep 50: ||h_L||=3.272e+04 ||g_2||=7.678e-07 eval_acc=0.3809 test_acc=0.3654
  ep 60: ||h_L||=3.598e+04 ||g_2||=7.403e-07 eval_acc=0.3799 test_acc=0.3669
  ep 70: ||h_L||=3.810e+04 ||g_2||=7.278e-07 eval_acc=0.3887 test_acc=0.3732
  ep 80: ||h_L||=3.930e+04 ||g_2||=7.065e-07 eval_acc=0.3857 test_acc=0.3711
  ep 90: ||h_L||=3.976e+04 ||g_2||=7.083e-07 eval_acc=0.3818 test_acc=0.3719
  ep 100: ||h_L||=3.983e+04 ||g_2||=7.093e-07 eval_acc=0.3818 test_acc=0.3721

FINAL test acc: 0.3721
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_lam0.001_s42.json