-improve

app.py

@@ -3,11 +3,13 @@ import pandas as pd
 import folium
 import numpy as np
 import os
+import json
 
 BASE        = os.path.dirname(os.path.abspath(__file__))
 STAY_POINTS = os.path.join(BASE, "data", "stay_points_sampled.csv")
 POI_PATH    = os.path.join(BASE, "data", "poi_sampled.csv")
 DEMO_PATH   = os.path.join(BASE, "data", "demographics_sampled.csv")
+COT_PATH    = os.path.join(BASE, "data", "cot_results.json")   # <-- your JSON file
 
 SEX_MAP  = {1:"Male", 2:"Female", -8:"Unknown", -7:"Prefer not to answer"}
 EDU_MAP  = {1:"Less than HS", 2:"HS Graduate/GED", 3:"Some College/Associate",
@@ -47,9 +49,184 @@ def parse_act_types(x):
 
 sp["act_label"] = sp["act_types"].apply(parse_act_types)
 
+# Load CoT JSON
+cot_by_agent = {}
+if os.path.exists(COT_PATH):
+    print("Loading CoT results...")
+    with open(COT_PATH, "r") as f:
+        cot_raw = json.load(f)
+    for result in cot_raw.get("inference_results", []):
+        cot_by_agent[result["agent_id"]] = result
+
+# All CSV agents are available; CoT will fall back to mock if no match
 sample_agents = sorted(sp["agent_id"].unique().tolist())
 print(f"Ready. {len(sample_agents)} agents loaded.")
 
+MOCK_COT = {
+    "text_representation": """MOBILITY TRAJECTORY DATA
+===========================
+Observation Period: 2024-01-29 to 2024-02-25 (28 days)
+Total Stay Points: 82
+Unique Locations: 16
+
+LOCATION PATTERNS
+----------------
+1. residence
+   Visits: 36 times
+   Average Duration: 829 minutes
+
+2. Clinton Mobile Estates
+   Visits: 9 times
+   Average Duration: 137 minutes
+
+3. 7-Eleven
+   Visits: 8 times
+   Average Duration: 118 minutes
+
+4. Euro Caffe
+   Visits: 5 times
+   Average Duration: 101 minutes
+
+5. Hoa Phap Hoa Dao Trang
+   Visits: 4 times
+   Average Duration: 112 minutes
+
+TEMPORAL PATTERNS
+----------------
+Activity by Time of Day:
+- morning: 56%
+- afternoon: 24%
+- evening: 20%
+
+Weekday vs Weekend:
+- weekday: 66%
+- weekend: 34%""",
+
+    "weekly_checkin": """WEEKLY CHECK-IN SUMMARY
+=======================
+Period: 2024-01-29 to 2024-02-25 (28 days)
+
+--- Monday, January 29 (Weekday) ---
+Total activities: 2
+- 09:10-10:14 (64 mins): Bear Wire - Work, Services, DropOff
+- 10:38-08:54 (1336 mins): residence - Home, Social Visit, DropOff
+
+--- Tuesday, January 30 (Weekday) ---
+Total activities: 2
+- 09:12-11:06 (115 mins): Clinton Mobile Estates - Work, Recreation, Exercise
+- 11:24-09:36 (1331 mins): residence - Home, Social Visit, DropOff
+
+--- Wednesday, January 31 (Weekday) ---
+Total activities: 4
+- 09:52-14:07 (255 mins): Castaway Cove Water Playground - Work, Exercise, DropOff
+- 14:21-15:27 (66 mins): residence - Home, Social Visit, DropOff
+- 15:44-19:03 (198 mins): Clinton Mobile Estates - Work, Recreation, Exercise
+- 19:21-09:06 (825 mins): residence - Home, Social Visit, DropOff
+
+--- Friday, February 02 (Weekday) ---
+Total activities: 3
+- 08:34-09:29 (55 mins): Euro Caffe - Work, Dining
+- 09:58-11:06 (68 mins): 7-Eleven - Work, Shopping, Dining, DropOff
+- 11:25-08:56 (1291 mins): residence - Home, Social Visit, DropOff
+
+--- Saturday, February 03 (Weekend) ---
+Total activities: 5
+- 09:13-10:03 (50 mins): Clinton Mobile Estates - Work, Recreation, Exercise
+- 10:27-11:10 (43 mins): 7-Eleven - Work, Shopping, Dining, DropOff
+- 11:29-18:39 (430 mins): residence - Home, Social Visit, DropOff
+- 18:56-21:30 (154 mins): Pepper Shaker Cafe - Work, Dining
+- 21:48-09:16 (689 mins): residence - Home, Social Visit, DropOff
+
+--- Sunday, February 11 (Weekend) ---
+Total activities: 5
+- 10:05-11:56 (111 mins): Hoa Phap Hoa Dao Trang - Religious
+- 12:21-13:36 (75 mins): Pepper Shaker Cafe - Work, Dining
+- 14:56-16:58 (122 mins): 7-Eleven - Work, Shopping, Dining, DropOff
+- 17:17-13:27 (1209 mins): residence - Home, Social Visit, DropOff
+
+WEEKLY INSIGHTS
+===============
+Most visited location: residence (36 visits)
+Time spent: DropOff 31.0% | Home 28.8% | Work 4.1% | Dining 2.3% | Exercise 1.7%
+Weekday avg activities: 2.7 | Weekend avg: 3.5""",
+
+    "step1_response": """Based on the provided mobility trajectory data, here are the objective features extracted:
+
+LOCATION INVENTORY:
+- POI categories visited:
+  - Residence
+  - Clinton Mobile Estates (Recreation, Exercise)
+  - 7-Eleven (Shopping, Dining, DropOff)
+  - Euro Caffe (Dining)
+  - Hoa Phap Hoa Dao Trang (Religious)
+  - Castaway Cove Water Playground (Exercise, DropOff)
+  - Salon 860 West (Shopping, Dining, DropOff)
+  - Pepper Shaker Cafe (Dining)
+  - Lucky Pierre Self Service Car Wash (Shopping, Dining, DropOff)
+
+- Frequency of visits:
+  - Residence: 36 times
+  - Clinton Mobile Estates: 9 times
+  - 7-Eleven: 8 times
+  - Euro Caffe: 5 times
+  - Hoa Phap Hoa Dao Trang: 4 times
+
+- Apparent price level:
+  - Budget: 7-Eleven, Lucky Pierre Self Service Car Wash
+  - Mid-range: Euro Caffe, Pepper Shaker Cafe
+
+TEMPORAL PATTERNS:
+- Active hours: 09:00 to 23:00
+- Weekly distribution: 66% weekday, 34% weekend
+- Regularity: Consistent daily routines
+- Duration: Longer stays at residence (829 min avg), shorter at other venues
+
+SPATIAL CHARACTERISTICS:
+- Activity radius: up to ~13 km from home
+- Geographic distribution: Urban areas, mix of residential and commercial
+
+SEQUENCE OBSERVATIONS:
+- Common transitions: Home→Clinton Mobile Estates→Home, Home→7-Eleven→Home
+- Weekday pattern: morning work/exercise, afternoon errands
+- Weekend pattern: morning religious visit, afternoon dining/shopping""",
+
+    "step2_response": """1. ROUTINE & SCHEDULE ANALYSIS:
+   - Semi-structured lifestyle with consistent morning start times (~09:00)
+   - Flexible schedule: varies between 2–5 activities per day
+   - Weekday/weekend contrast suggests work-centric weekdays, more leisure on weekends
+
+2. ECONOMIC BEHAVIOR PATTERNS:
+   - Mix of budget-conscious and mid-range spending
+   - Frequent 7-Eleven visits suggest convenience-oriented, cost-aware shopping
+   - No signs of premium or luxury venue preferences
+   - Activity radius (~13 km) suggests personal vehicle access
+
+3. SOCIAL & LIFESTYLE INDICATORS:
+   - Regular weekend religious attendance (Hoa Phap Hoa Dao Trang) → community ties
+   - Recreational activities (Castaway Cove, Clinton Mobile Estates) → active lifestyle
+   - Dining out occasionally (Euro Caffe, Pepper Shaker Cafe) → moderate social life
+   - High home-time suggests family-oriented or home-centered lifestyle
+
+4. URBAN LIFESTYLE CHARACTERISTICS:
+   - Urban/suburban mix: lives in residential area, travels to nearby commercial zones
+   - Activity radius and venue diversity suggest working-class to middle-class area
+   - Community engagement evident through religious and recreational venues
+
+5. ROUTINE STABILITY:
+   - Highly consistent patterns over 4 weeks with minimal deviation
+   - No signs of major life transitions
+   - Regular work-like attendance at Clinton Mobile Estates suggests stable employment""",
+
+    "step3_response": """INCOME_PREDICTION: Middle ($35k-$75k)
+INCOME_CONFIDENCE: 4
+INCOME_REASONING: The individual's mobility patterns suggest a mix of budget-conscious and mid-range spending. Frequent visits to 7-Eleven and self-service venues indicate cost awareness, while occasional mid-range dining (Euro Caffe, Pepper Shaker Cafe) suggests some disposable income. The neighborhood types are commercial/residential mixed, consistent with a middle-class area. No luxury venue visits detected. Activity radius of ~13 km and apparent vehicle access align with middle-income transport patterns.
+
+RANKED ALTERNATIVES:
+1. Middle ($35k-$75k) — Primary prediction
+2. Low ($15k-$35k) — Budget venue frequency could suggest lower income
+3. Upper-Middle ($75k-$125k) — Unlikely given absence of premium venues"""
+}
+
 
 def build_map(agent_sp):
     agent_sp = agent_sp.reset_index(drop=True).copy()
@@ -62,14 +239,21 @@ def build_map(agent_sp):
 
     coords = list(zip(agent_sp["latitude"], agent_sp["longitude"]))
     if len(coords) > 1:
-        folium.PolyLine(coords, color="#aaaaaa", weight=1.5, opacity=0.4).add_to(m)
+        folium.PolyLine(coords, color="#cc000055", weight=1.5, opacity=0.4).add_to(m)
 
     n = len(agent_sp)
     for i, row in agent_sp.iterrows():
+        # Red gradient: light red (#ffcccc) → deep red (#8b0000)
         ratio = i / max(n - 1, 1)
-        r = int(255 * ratio)
-        g = int(255 * (1 - ratio))
-        color = f"#{r:02x}{g:02x}33"
+        r = 255
+        g = int(204 * (1 - ratio))   # 204 → 0
+        b = int(204 * (1 - ratio))   # 204 → 0
+        # Clamp deep end toward dark red (139, 0, 0)
+        r = int(255 - ratio * (255 - 139))  # 255 → 139
+        g = int(204 * (1 - ratio) * (1 - ratio * 0.3))  # fade to 0
+        b = 0
+        color = f"#{r:02x}{g:02x}{b:02x}"
+
         folium.CircleMarker(
             location=[row["latitude"], row["longitude"]],
             radius=7, color=color, fill=True, fill_color=color, fill_opacity=0.9,
@@ -83,21 +267,10 @@ def build_map(agent_sp):
         ).add_to(m)
 
     m.get_root().width = "100%"
-    m.get_root().height = "500px"
+    m.get_root().height = "420px"
     return m._repr_html_()
 
 
-def build_poi_sequence(agent_sp):
-    lines = []
-    for _, row in agent_sp.iterrows():
-        lines.append(
-            f"{row['start_datetime'].strftime('%a %m/%d')}  "
-            f"{row['start_datetime'].strftime('%H:%M')}–{row['end_datetime'].strftime('%H:%M')}  "
-            f"({int(row['duration_min'])} min)  |  {row['name']}  |  {row['act_label']}"
-        )
-    return "\n".join(lines)
-
-
 def build_demo_text(row):
     age  = int(row["age"]) if row["age"] > 0 else "Unknown"
     return (
@@ -109,26 +282,115 @@ def build_demo_text(row):
     )
 
 
+def parse_step3(text):
+    """Extract prediction, confidence, reasoning from step3 response text."""
+    prediction, confidence, reasoning = "", "", ""
+    for line in text.splitlines():
+        line = line.strip()
+        if line.startswith("INCOME_PREDICTION:"):
+            prediction = line.replace("INCOME_PREDICTION:", "").strip()
+        elif line.startswith("INCOME_CONFIDENCE:"):
+            confidence = line.replace("INCOME_CONFIDENCE:", "").strip()
+        elif line.startswith("INCOME_REASONING:"):
+            reasoning = line.replace("INCOME_REASONING:", "").strip()
+    return prediction, confidence, reasoning
+
+
 def on_select(agent_id):
     agent_id   = int(agent_id)
     agent_sp   = sp[sp["agent_id"] == agent_id].sort_values("start_datetime")
     agent_demo = demo[demo["agent_id"] == agent_id].iloc[0]
-    return build_map(agent_sp), build_poi_sequence(agent_sp), build_demo_text(agent_demo)
+    cot        = cot_by_agent.get(agent_id, MOCK_COT)
+
+    map_html  = build_map(agent_sp)
+    demo_text = build_demo_text(agent_demo)
+
+    # NUMOSIM raw data
+    raw_text = cot.get("text_representation", "") + "\n\n" + cot.get("weekly_checkin", "")
+
+    # CoT stages
+    step1 = cot.get("step1_response", "No data")
+    step2 = cot.get("step2_response", "No data")
+    step3_raw = cot.get("step3_response", "No data")
+    pred, conf, reason = parse_step3(step3_raw)
+    step3_summary = f"INCOME PREDICTION: {pred}\nCONFIDENCE: {conf}/5\n\nREASONING:\n{reason}\n\n---FULL RESPONSE---\n{step3_raw}"
+
+    return map_html, raw_text, step1, step2, step3_summary, demo_text
 
 
-with gr.Blocks(title="HiCoTraj Demo", theme=gr.themes.Soft()) as app:
-    gr.Markdown("## HiCoTraj: Trajectory Visualization")
+custom_css = """
+.gradio-container { max-width: 1600px !important; }
+.stage-label { font-weight: bold; color: #b22222; }
+"""
 
+with gr.Blocks(title="HiCoTraj Demo", theme=gr.themes.Soft(), css=custom_css) as app:
+    gr.Markdown("## 🗺️ HiCoTraj: Trajectory Visualization & Chain-of-Thought Demo")
+    gr.Markdown("*Zero-Shot Demographic Reasoning via Hierarchical Chain-of-Thought Prompting from Trajectory*")
+
+    # ── Top bar ──────────────────────────────────────────────────────────────
     with gr.Row():
-        agent_dd   = gr.Dropdown(choices=[str(a) for a in sample_agents],
-                                 label="Select Agent", value=str(sample_agents[0]))
-        demo_label = gr.Textbox(label="Ground Truth Demographics", interactive=False)
+        agent_dd   = gr.Dropdown(
+            choices=[str(a) for a in sample_agents],
+            label="Select Agent",
+            value=str(sample_agents[0]),
+            scale=1
+        )
+        demo_label = gr.Textbox(
+            label="Ground Truth Demographics",
+            interactive=False,
+            scale=4
+        )
 
-    map_out = gr.HTML(label="Trajectory Map")
-    poi_out = gr.Textbox(label="POI Sequence", lines=20, interactive=False)
+    # ── Main content: Left | Right ────────────────────────────────────────
+    with gr.Row():
+
+        # LEFT: Map + NUMOSIM raw data
+        with gr.Column(scale=1):
+            gr.Markdown("### 📍 Trajectory Map")
+            map_out = gr.HTML(label="Trajectory Map")
+
+            gr.Markdown("### 📋 NUMOSIM Raw Data")
+            raw_out = gr.Textbox(
+                label="Mobility Summary + Weekly Check-in",
+                lines=25,
+                interactive=False
+            )
 
-    agent_dd.change(fn=on_select, inputs=agent_dd, outputs=[map_out, poi_out, demo_label])
-    app.load(fn=on_select, inputs=agent_dd, outputs=[map_out, poi_out, demo_label])
+        # RIGHT: CoT three stages
+        with gr.Column(scale=1):
+            gr.Markdown("### 🧠 Hierarchical Chain-of-Thought Reasoning")
+
+            with gr.Accordion("📌 Stage 1 — Factual Feature Extraction", open=True):
+                step1_out = gr.Textbox(
+                    label="Stage 1 Response",
+                    lines=12,
+                    interactive=False
+                )
+
+            with gr.Accordion("🔍 Stage 2 — Behavioral Pattern Analysis", open=True):
+                step2_out = gr.Textbox(
+                    label="Stage 2 Response",
+                    lines=12,
+                    interactive=False
+                )
+
+            with gr.Accordion("🎯 Stage 3 — Demographic Inference", open=True):
+                step3_out = gr.Textbox(
+                    label="Stage 3 Response (Income Prediction)",
+                    lines=12,
+                    interactive=False
+                )
+
+    agent_dd.change(
+        fn=on_select,
+        inputs=agent_dd,
+        outputs=[map_out, raw_out, step1_out, step2_out, step3_out, demo_label]
+    )
+    app.load(
+        fn=on_select,
+        inputs=agent_dd,
+        outputs=[map_out, raw_out, step1_out, step2_out, step3_out, demo_label]
+    )
 
 if __name__ == "__main__":
     app.launch(share=True)