======================================================================== PREDICTOR COMPARISON: BNNBot (Linear+Hebbian) vs WiSARD K=12 vs TsetlinBot (RTM) Rep power: p1.07 | Hit threshold: 18.0 (0-99 scale, ~18px) Online learning, cumulative per-enemy | Scores measured across all powers ======================================================================== Enemy: crazy battles=5 rows=2908 Avg MAE (all powers): BNNBot=24.72 WiSARD=22.29 TsetlinBot=27.10 (TM-warm=20.98) Avg Hit% (all powers): BNNBot= 46.0% WiSARD= 51.4% TsetlinBot= 38.7% Winner by MAE: WiSARD Per-power breakdown (MAE, crazy): Power BNNBot WiSARD TsetlinBot Best ---------------------------------------------------- p0.10 17.14 15.09 19.44 WiSARD p0.42 18.17 16.03 20.56 WiSARD p0.74 19.39 17.13 21.83 WiSARD p1.07 20.86 18.52 23.33 WiSARD p1.39 22.52 20.09 25.00 WiSARD p1.71 24.44 21.93 26.92 WiSARD p2.03 26.71 24.13 29.16 WiSARD p2.36 29.39 26.77 31.78 WiSARD p2.68 32.47 29.81 34.77 WiSARD p3.00 36.07 33.43 38.25 WiSARD Enemy: spinbot battles=5 rows=1163 Avg MAE (all powers): BNNBot=23.75 WiSARD=24.54 TsetlinBot=30.20 (TM-warm=29.11) Avg Hit% (all powers): BNNBot= 43.7% WiSARD= 41.1% TsetlinBot= 37.1% Winner by MAE: BNNBot(Lin+Heb) Per-power breakdown (MAE, spinbot): Power BNNBot WiSARD TsetlinBot Best ---------------------------------------------------- p0.10 17.81 18.12 23.61 BNNBot p0.42 18.60 19.11 24.83 BNNBot p0.74 19.60 20.24 26.14 BNNBot p1.07 20.78 21.55 27.52 BNNBot p1.39 22.13 22.98 28.92 BNNBot p1.71 23.68 24.60 30.42 BNNBot p2.03 25.44 26.40 32.04 BNNBot p2.36 27.45 28.43 33.87 BNNBot p2.68 29.69 30.67 35.99 BNNBot p3.00 32.33 33.28 38.64 BNNBot Enemy: target battles=5 rows=1606 Avg MAE (all powers): BNNBot= 2.68 WiSARD= 2.21 TsetlinBot= 4.14 (TM-warm= 6.50) Avg Hit% (all powers): BNNBot= 95.5% WiSARD= 97.0% TsetlinBot= 93.5% Winner by MAE: WiSARD Per-power breakdown (MAE, target): Power BNNBot WiSARD TsetlinBot Best ---------------------------------------------------- p0.10 1.71 1.61 3.34 WiSARD p0.42 1.83 1.62 3.44 WiSARD p0.74 1.97 1.67 3.57 WiSARD p1.07 2.16 1.76 3.72 WiSARD p1.39 2.37 1.88 3.88 WiSARD p1.71 2.61 2.06 4.08 WiSARD p2.03 2.92 2.30 4.32 WiSARD p2.36 3.27 2.60 4.61 WiSARD p2.68 3.71 3.02 4.99 WiSARD p3.00 4.23 3.55 5.45 WiSARD Enemy: walls battles=5 rows=1661 Avg MAE (all powers): BNNBot=15.31 WiSARD=13.22 TsetlinBot=18.81 (TM-warm=15.17) Avg Hit% (all powers): BNNBot= 69.8% WiSARD= 76.2% TsetlinBot= 62.4% Winner by MAE: WiSARD Per-power breakdown (MAE, walls): Power BNNBot WiSARD TsetlinBot Best ---------------------------------------------------- p0.10 10.38 8.81 13.27 WiSARD p0.42 10.90 9.21 14.03 WiSARD p0.74 11.60 9.77 14.93 WiSARD p1.07 12.43 10.47 15.94 WiSARD p1.39 13.51 11.42 17.12 WiSARD p1.71 14.81 12.61 18.50 WiSARD p2.03 16.40 14.12 20.15 WiSARD p2.36 18.37 16.00 22.13 WiSARD p2.68 20.83 18.41 24.54 WiSARD p3.00 23.85 21.41 27.51 WiSARD ======================================================================== OVERALL SUMMARY (all enemies, all powers combined) ======================================================================== n=73380 data points MAE: BNNBot=17.61 WiSARD=16.20 TsetlinBot=20.69 (TM warm-phase=18.77) Hit%: BNNBot=61.9% WiSARD=65.4% TsetlinBot=55.8% Wins per enemy type (by avg MAE, all powers): WiSARD: 3/4 BNNBot: 1/4 VERDICT (all-rows MAE): WiSARD K=12 VERDICT (TM warm-phase): WiSARD K=12 Notes: - BNNBot: linear extrapolation + 24-cell Hebbian residual table (lr=0.1) - WiSARD: K=12, 276-bit input (4 frames × 69 bits), 23 LUT nodes, bleach=1 - TsetlinBot (RTM): 60 clauses, N_states=15, s=3.0, T=30, shared x/y RTMs - TM warm = last-50-rows MAE; accounts for cold-start overhead - Hit threshold: 18.0 (0-99 scale units, ~equivalent to tank body) - Residuals learned on mean across all powers; tested at each power level - Enemy types: crazy, spinbot, target, walls