File size: 12,326 Bytes
1ceda33 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 |
#!/usr/bin/env python3
"""
Figure 5: Human-AI Session Comparison
Generates a 3-panel comparison of:
- Session A (Deep Dive): Baseline, no metrics shown
- Session B (Whiplash): Real metrics shown, topic switches
- Session C (Gaslight): Adversarial metrics injected
Usage:
python 15_human_session_comparison.py
Output:
figures/FIG5_human_session_comparison.png
"""
import json
import sys
from pathlib import Path
from typing import Any, Dict, List, Optional, Tuple
import matplotlib.pyplot as plt
import matplotlib.patches as mpatches
import numpy as np
# Paths
SCRIPT_DIR = Path(__file__).parent
RESULTS_DIR = SCRIPT_DIR.parent / "results"
FIGURES_DIR = SCRIPT_DIR.parent.parent / "figures"
# Session files (update if filenames change)
SESSION_FILES = {
"A_deep": RESULTS_DIR / "human_run_01_deep.json",
"B_whiplash": RESULTS_DIR / "09_steering_2026-01-19_22-58-59_human_session.json",
"C_gaslight": RESULTS_DIR / "09_steering_2026-01-20_13-27-51_human_session.json",
}
SESSION_LABELS = {
"A_deep": "Session A: Deep Dive\n(Baseline - No metrics shown)",
"B_whiplash": "Session B: Topic Switching\n(Real metrics shown)",
"C_gaslight": "Session C: Metrics Spoofing\n(Adversarial telemetry injected)",
}
SESSION_COLORS = {
"A_deep": "#3498db", # Blue
"B_whiplash": "#9b59b6", # Purple
"C_gaslight": "#e74c3c", # Red
}
STATE_COLORS = {
"stable": "#2ecc71", # Green
"protostar": "#f39c12", # Orange
"split": "#e74c3c", # Red
}
def load_session(filepath: Path) -> Tuple[List[Dict], str]:
"""Load session data and return turns + condition name."""
with open(filepath, 'r', encoding='utf-8') as f:
data = json.load(f)
# Handle different formats
if isinstance(data, dict) and "results" in data:
results = data["results"]
if results and len(results) > 0:
return results[0].get("turns", []), results[0].get("condition_name", "unknown")
elif isinstance(data, dict) and "turns" in data:
return data["turns"], data.get("condition", "unknown")
elif isinstance(data, list):
# Old format: list of condition results
if len(data) > 0:
return data[0].get("turns", []), data[0].get("condition_name", "unknown")
return [], "unknown"
def extract_metrics(turns: List[Dict]) -> Dict[str, List]:
"""Extract time series from turns."""
xs = []
sgi_vals = []
vel_vals = []
ctx_states = []
ctx_ids = []
for i, turn in enumerate(turns, start=1):
xs.append(i)
real = turn.get("real_metrics", {})
# Get SGI (prefer turn_pair_sgi)
sgi = (
real.get("turn_pair_sgi_latest") or
real.get("turn_pair_sgi_mean") or
real.get("sgi_latest") or
real.get("sgi_mean")
)
sgi_vals.append(float(sgi) if sgi is not None else None)
# Get Velocity (prefer orbital)
vel = (
real.get("orbital_velocity_latest") or
real.get("orbital_velocity_mean") or
real.get("velocity_latest") or
real.get("velocity_mean")
)
vel_vals.append(float(vel) if vel is not None else None)
# Context state (last in array or latest)
state_arr = real.get("per_turn_context_state", [])
if state_arr:
ctx_states.append(state_arr[-1] if state_arr else "stable")
else:
ctx_states.append(real.get("context_state_latest", "stable") or "stable")
# Context ID
id_arr = real.get("per_turn_context_id", [])
if id_arr:
ctx_ids.append(id_arr[-1] if id_arr else "ctx_1")
else:
ctx_ids.append(real.get("context_id_latest", "ctx_1") or "ctx_1")
return {
"xs": xs,
"sgi": sgi_vals,
"velocity": vel_vals,
"context_state": ctx_states,
"context_id": ctx_ids,
}
def compute_stats(metrics: Dict[str, List]) -> Dict[str, Any]:
"""Compute summary statistics."""
sgi_clean = [v for v in metrics["sgi"] if v is not None]
vel_clean = [v for v in metrics["velocity"] if v is not None]
# Count context states
state_counts = {}
for s in metrics["context_state"]:
state_counts[s] = state_counts.get(s, 0) + 1
# Count context switches
ctx_switches = 0
prev_ctx = None
for ctx in metrics["context_id"]:
if prev_ctx is not None and ctx != prev_ctx:
ctx_switches += 1
prev_ctx = ctx
return {
"sgi_mean": np.mean(sgi_clean) if sgi_clean else None,
"sgi_std": np.std(sgi_clean) if sgi_clean else None,
"vel_mean": np.mean(vel_clean) if vel_clean else None,
"vel_std": np.std(vel_clean) if vel_clean else None,
"state_counts": state_counts,
"context_switches": ctx_switches,
"n_turns": len(metrics["xs"]),
}
def plot_comparison():
"""Generate the 3-panel comparison figure."""
# Load all sessions
sessions = {}
for key, filepath in SESSION_FILES.items():
if filepath.exists():
turns, condition = load_session(filepath)
metrics = extract_metrics(turns)
stats = compute_stats(metrics)
sessions[key] = {
"turns": turns,
"condition": condition,
"metrics": metrics,
"stats": stats,
}
print(f"[OK] Loaded {key}: {len(turns)} turns")
else:
print(f"[WARN] File not found: {filepath}")
if len(sessions) < 3:
print(f"[WARN] Only {len(sessions)}/3 sessions found")
# Create figure: 3 columns x 3 rows
# Row 1: SGI over time
# Row 2: Velocity over time
# Row 3: Phase space (SGI x Velocity)
fig, axes = plt.subplots(3, 3, figsize=(18, 14))
session_keys = ["A_deep", "B_whiplash", "C_gaslight"]
for col, key in enumerate(session_keys):
if key not in sessions:
for row in range(3):
axes[row, col].set_visible(False)
continue
data = sessions[key]
m = data["metrics"]
stats = data["stats"]
color = SESSION_COLORS[key]
xs = m["xs"]
sgi = [v if v is not None else np.nan for v in m["sgi"]]
vel = [v if v is not None else np.nan for v in m["velocity"]]
point_colors = [STATE_COLORS.get(s, "#3498db") for s in m["context_state"]]
# Row 1: SGI
ax1 = axes[0, col]
ax1.plot(xs, sgi, linewidth=2, color=color, alpha=0.7)
ax1.scatter(xs, sgi, c=point_colors, s=60, zorder=5, edgecolors='white', linewidth=0.5)
ax1.axhline(1.0, color="gray", linestyle="--", alpha=0.6)
ax1.set_ylabel("SGI (Orbital Radius)" if col == 0 else "")
ax1.set_ylim(0, 2.0)
ax1.set_title(SESSION_LABELS[key], fontweight="bold", fontsize=11)
ax1.grid(True, alpha=0.3)
# Add stats annotation
if stats["sgi_mean"] is not None:
ax1.text(0.95, 0.95, f"mean={stats['sgi_mean']:.2f}\nstd={stats['sgi_std']:.2f}",
transform=ax1.transAxes, ha='right', va='top', fontsize=9,
bbox=dict(boxstyle='round', facecolor='white', alpha=0.8))
# Row 2: Velocity
ax2 = axes[1, col]
ax2.plot(xs, vel, linewidth=2, color=color, alpha=0.7)
ax2.scatter(xs, vel, c=point_colors, s=60, zorder=5, edgecolors='white', linewidth=0.5)
ax2.set_ylabel("Velocity (degrees)" if col == 0 else "")
ax2.set_xlabel("Turn")
ax2.set_ylim(0, 180)
ax2.grid(True, alpha=0.3)
# Add stats annotation
if stats["vel_mean"] is not None:
ax2.text(0.95, 0.95, f"mean={stats['vel_mean']:.1f}deg\nstd={stats['vel_std']:.1f}deg",
transform=ax2.transAxes, ha='right', va='top', fontsize=9,
bbox=dict(boxstyle='round', facecolor='white', alpha=0.8))
# Row 3: Phase space
ax3 = axes[2, col]
# Coherence region
sgi_min, sgi_max = 0.7, 1.3
vel_min, vel_max = 15, 45
ax3.add_patch(
plt.Rectangle(
(sgi_min, vel_min),
sgi_max - sgi_min,
vel_max - vel_min,
facecolor="#2ecc71",
alpha=0.15,
edgecolor="#2ecc71",
linewidth=2,
linestyle="--",
)
)
ax3.axvline(1.0, color="gray", linestyle="--", alpha=0.6, linewidth=1)
ax3.plot(1.0, 30, "g*", markersize=14, zorder=5)
# Plot trajectory with time gradient
# IMPORTANT: build paired points to keep x/y/color lengths identical
phase_points = [
(sgi_v, vel_v, STATE_COLORS.get(state, "#3498db"))
for sgi_v, vel_v, state in zip(m["sgi"], m["velocity"], m["context_state"])
if sgi_v is not None and vel_v is not None
]
phase_sgi = [p[0] for p in phase_points]
phase_vel = [p[1] for p in phase_points]
phase_colors = [p[2] for p in phase_points]
if len(phase_points) > 1:
# Time gradient
cmap = plt.cm.viridis
for i in range(len(phase_sgi) - 1):
ax3.plot(
[phase_sgi[i], phase_sgi[i+1]],
[phase_vel[i], phase_vel[i+1]],
color=cmap(i / len(phase_sgi)),
linewidth=1.5,
alpha=0.6
)
# Scatter with state colors
ax3.scatter(phase_sgi, phase_vel, c=phase_colors, s=80, zorder=5, edgecolors='white', linewidth=0.5)
# Mark start and end
ax3.scatter([phase_sgi[0]], [phase_vel[0]], marker='s', s=120, c='blue', zorder=6, label='Start')
ax3.scatter([phase_sgi[-1]], [phase_vel[-1]], marker='^', s=120, c='red', zorder=6, label='End')
ax3.set_xlabel("SGI (Orbital Radius)")
ax3.set_ylabel("Velocity (degrees)" if col == 0 else "")
ax3.set_xlim(0, 2.0)
ax3.set_ylim(0, 180)
ax3.grid(True, alpha=0.3)
# Context switch annotation
ax3.text(0.05, 0.95, f"Context Switches: {stats['context_switches']}",
transform=ax3.transAxes, ha='left', va='top', fontsize=9,
bbox=dict(boxstyle='round', facecolor='white', alpha=0.8))
# Add legend for context states
legend_patches = [
mpatches.Patch(color=STATE_COLORS["stable"], label="Stable (anchored)"),
mpatches.Patch(color=STATE_COLORS["protostar"], label="Protostar (forming)"),
mpatches.Patch(color=STATE_COLORS["split"], label="Split (context change)"),
]
fig.legend(handles=legend_patches, loc='upper center', ncol=3,
bbox_to_anchor=(0.5, 0.02), fontsize=10)
# Main title
fig.suptitle("Figure 5: Human-AI Session Comparison\nSemantic Dynamics Across Conditions",
fontweight="bold", fontsize=14, y=0.98)
plt.tight_layout(rect=[0, 0.05, 1, 0.95])
# Save
FIGURES_DIR.mkdir(parents=True, exist_ok=True)
output_path = FIGURES_DIR / "FIG5_human_session_comparison.png"
plt.savefig(output_path, dpi=200, bbox_inches="tight", facecolor="white")
print(f"\n[OK] Saved: {output_path}")
# Also save individual stats
print("\n" + "="*70)
print("SESSION COMPARISON SUMMARY")
print("="*70)
for key in session_keys:
if key not in sessions:
continue
stats = sessions[key]["stats"]
print(f"\n{SESSION_LABELS[key].split(chr(10))[0]}:")
print(f" Turns: {stats['n_turns']}")
if stats["sgi_mean"]:
print(f" SGI: {stats['sgi_mean']:.3f} +/- {stats['sgi_std']:.3f}")
if stats["vel_mean"]:
print(f" Velocity: {stats['vel_mean']:.1f} +/- {stats['vel_std']:.1f} deg")
print(f" Context Switches: {stats['context_switches']}")
print(f" States: {stats['state_counts']}")
print("\n" + "="*70)
plt.show()
if __name__ == "__main__":
plot_comparison()
|