cot

app.py

@@ -285,75 +285,103 @@ def _dots():
 
 
 def render_chain(s1_text, s2_text, s3_text, status="done"):
-    # status: idle | running1 | running2 | running3 | done
     s1_active = status in ("running1", "running2", "running3", "done")
     s2_active = status in ("running2", "running3", "done")
     s3_active = status in ("running3", "done")
 
-    # ── Stage 1 ───────────────────────────────────────────────────────────────
+    # ── Stage 1: extract bullet lines from free text ──────────────────────────
     if status == "running1":
         s1_content = f'<div class="thinking">Extracting features {_dots()}</div>'
     elif s1_text:
         tags = []
         for line in s1_text.splitlines():
             line = line.strip().lstrip("-").strip()
-            if line and len(line) < 65 and not line.endswith(":"):
+            # Keep lines that look like "Location: N visits, description"
+            if line and len(line) > 8 and len(line) < 80 and not line.endswith(":"):
                 tags.append(line)
-            if len(tags) >= 9:
+            if len(tags) >= 8:
                 break
-        s1_content = '<div class="tag-row">' + \
-            "".join(f'<span class="tag">{t}</span>' for t in tags[:9]) + \
-            '</div>'
+        s1_content = '<div class="tag-row">' + "".join(f'<span class="tag">{t}</span>' for t in tags) + '</div>'
     else:
-        s1_content = '<div class="empty-hint">Press ▶ Run HiCoTraj to start</div>'
-
-    # ── Stage 2 ───────────────────────────────────────────────────────────────
-    KEYS = ["SCHEDULE", "ECONOMIC", "SOCIAL", "LIFESTYLE", "STABILITY"]
+        s1_content = '<div class="empty-hint">Press ▶ to start</div>'
+
+    # ── Stage 2: extract numbered sections as key-value rows ──────────────────
+    KEYS = [
+        ("SCHEDULE",  ["ROUTINE", "SCHEDULE"]),
+        ("ECONOMIC",  ["ECONOMIC", "SPENDING", "FINANCIAL"]),
+        ("SOCIAL",    ["SOCIAL", "COMMUNITY"]),
+        ("LIFESTYLE", ["LIFESTYLE", "ACTIVITY", "LEISURE"]),
+        ("STABILITY", ["STABILITY", "CONSISTENCY", "PATTERN"]),
+    ]
     if status == "running2":
         s2_content = f'<div class="thinking" style="color:#a06030">Analyzing behavior {_dots()}</div>'
     elif s2_text:
         rows_html = ""
-        for key in KEYS:
-            m = re.search(rf"{key}[:\s]+(.+)", s2_text, re.IGNORECASE)
-            val = m.group(1).strip().rstrip(".") if m else "—"
-            if len(val) > 85:
-                val = val[:82] + "..."
-            rows_html += f'<div class="bkey">{key}</div><div class="bval">{val}</div>'
+        lines = s2_text.splitlines()
+        # Build a searchable blob per section by scanning numbered headings
+        sections = {}
+        current_key = None
+        current_lines = []
+        for line in lines:
+            line = line.strip()
+            # Detect numbered headings like "1. ROUTINE & SCHEDULE ANALYSIS:"
+            heading_match = re.match(r'^\d+\.\s+(.+?):\s*$', line)
+            if heading_match:
+                if current_key:
+                    sections[current_key] = " ".join(current_lines)
+                current_key = heading_match.group(1).upper()
+                current_lines = []
+            elif current_key and line.startswith("-"):
+                current_lines.append(line.lstrip("-").strip())
+        if current_key:
+            sections[current_key] = " ".join(current_lines)
+
+        for label, search_words in KEYS:
+            val = "—"
+            for k, v in sections.items():
+                if any(w in k for w in search_words):
+                    # Take first sentence
+                    first = v.split(".")[0].strip() if v else ""
+                    val = first[:90] + ("..." if len(first) > 90 else "")
+                    break
+            rows_html += f'<div class="bkey">{label}</div><div class="bval">{val}</div>'
         s2_content = f'<div class="behavior-row">{rows_html}</div>'
     else:
         s2_content = '<div class="empty-hint">Waiting...</div>'
 
-    # ── Stage 3 ───────────────────────────────────────────────────────────────
+    # ── Stage 3: prediction ───────────────────────────────────────────────────
     if status == "running3":
         s3_content = f'<div class="thinking" style="color:#c0392b">Inferring demographics {_dots()}</div>'
     elif s3_text:
-        pred = conf_raw = reasoning = alts = ""
+        pred = reasoning = alts = ""
         for line in s3_text.splitlines():
             line = line.strip()
             if line.startswith("INCOME_PREDICTION:"):
                 pred = line.replace("INCOME_PREDICTION:", "").strip()
-            elif line.startswith("INCOME_CONFIDENCE:"):
-                conf_raw = line.replace("INCOME_CONFIDENCE:", "").strip()
             elif line.startswith("INCOME_REASONING:"):
                 reasoning = line.replace("INCOME_REASONING:", "").strip()
             elif line.startswith("ALTERNATIVES:"):
                 alts = line.replace("ALTERNATIVES:", "").strip()
-        try:
-            conf_int = int(re.search(r"\d", conf_raw).group())
-        except:
-            conf_int = 3
-        bar_pct = conf_int * 20
+        # If reasoning spans multiple lines, grab all of it
+        if not reasoning:
+            in_reasoning = False
+            reasoning_parts = []
+            for line in s3_text.splitlines():
+                line = line.strip()
+                if line.startswith("INCOME_REASONING:"):
+                    in_reasoning = True
+                    reasoning_parts.append(line.replace("INCOME_REASONING:", "").strip())
+                elif in_reasoning and line and not line.startswith("INCOME_") and not line.startswith("ALTERNATIVES"):
+                    reasoning_parts.append(line)
+                elif in_reasoning and (line.startswith("INCOME_") or line.startswith("ALTERNATIVES")):
+                    break
+            reasoning = " ".join(reasoning_parts).strip()
+
         alts_html = f'<div class="alternatives">Also possible: <span>{alts}</span></div>' if alts else ""
         s3_content = f"""
         <div class="pred-block">
             <div class="pred-label">Income Prediction</div>
             <div class="pred-value">{pred or "—"}</div>
-            <div class="confidence-bar-wrap">
-                <div class="confidence-bar-bg">
-                    <div class="confidence-bar-fill" style="width:{bar_pct}%"></div>
-                </div>
-                <div class="confidence-label">Confidence {conf_int}/5</div>
-            </div>
             <div class="reasoning-text">{reasoning}</div>
             {alts_html}
         </div>"""
@@ -362,14 +390,15 @@ def render_chain(s1_text, s2_text, s3_text, status="done"):
 
     PROMPT_SNIPPETS = {
         "s1": "You are an expert mobility analyst. Given the trajectory data below, extract: (1) LOCATION INVENTORY — list all POI categories visited and visit frequency; (2) TEMPORAL PATTERNS — weekly distribution, peak hours; (3) SEQUENCE — typical activity chains...",
-        "s2": "Based on the trajectory features identified: {Response 1}. Now analyze what these mobility patterns reveal about lifestyle: (1) SCHEDULE — work/activity routine type; (2) ECONOMIC — spending venue tiers; (3) SOCIAL — social engagement patterns; (4) LIFESTYLE — activity diversity; (5) STABILITY — consistency of routine...",
-        "s3": "Based on feature analysis {Response 1} and behavioral analysis {Response 2}, predict income level. Output format — INCOME_PREDICTION: [Low (<$25k) | Lower-middle ($25k–$50k) | Middle ($50k–$100k) | Upper-middle ($100k–$150k) | High (>$150k)]; INCOME_CONFIDENCE: [1–5]; INCOME_REASONING: [detailed reasoning]...",
+        "s2": "Based on the trajectory features identified: {Response 1}. Now analyze what these mobility patterns reveal about lifestyle: (1) SCHEDULE — work/activity routine type; (2) ECONOMIC — spending venue tiers; (3) SOCIAL — social engagement; (4) LIFESTYLE — activity diversity; (5) STABILITY — consistency of routine...",
+        "s3": "Based on feature analysis {Response 1} and behavioral analysis {Response 2}, predict income level. Output — INCOME_PREDICTION: [range]; INCOME_REASONING: [detailed reasoning]...",
     }
-    def card(cls, badge, title, content, active):
+
+    def card(cls, badge, title, content_html, active):
         dim = "active" if active else "dim"
         prompt = PROMPT_SNIPPETS.get(cls, "")
-        prompt_html = f'''<div class="prompt-snippet"><span class="prompt-label">Prompt</span>{prompt}</div>''' if prompt else ""
-        resp_label = '<div class="resp-label">Response</div>' if active and content and "empty-hint" not in content and "thinking" not in content else ""
+        prompt_html = f'<div class="prompt-snippet"><span class="prompt-label">Prompt</span>{prompt}</div>' if prompt else ""
+        resp_label = '<div class="resp-label">Response</div>' if active and content_html and "empty-hint" not in content_html and "thinking" not in content_html else ""
         return f"""
         <div class="stage-card {cls} {dim}">
             <div class="stage-header">
@@ -378,7 +407,7 @@ def render_chain(s1_text, s2_text, s3_text, status="done"):
             </div>
             {prompt_html}
             {resp_label}
-            {content}
+            {content_html}
         </div>"""
 
     def arrow(label, active):
@@ -571,8 +600,8 @@ with gr.Blocks(title="HiCoTraj Demo", theme=gr.themes.Soft()) as app:
     gr.Markdown("## HiCoTraj — Trajectory Visualization & Hierarchical CoT Demo")
     gr.Markdown("*Zero-Shot Demographic Reasoning via Hierarchical Chain-of-Thought Prompting from Trajectory*  ·  ACM SIGSPATIAL GeoGenAgent 2025")
     gr.Markdown("""
-**Dataset:** NUMOSIM[1]
-> [1]Stanford C, Adari S, Liao X, et al. *NUMoSim: A Synthetic Mobility Dataset with Anomaly Detection Benchmarks.* ACM SIGSPATIAL Workshop on Geospatial Anomaly Detection, 2024.
+**Dataset:** NUMOSIM — a synthetic mobility dataset with realistic activity patterns across 6,000 agents.
+> Stanford C, Adari S, Liao X, et al. *NUMoSim: A Synthetic Mobility Dataset with Anomaly Detection Benchmarks.* ACM SIGSPATIAL Workshop on Geospatial Anomaly Detection, 2024.
 """)
 
     gr.Markdown("### Select Agent")