File size: 10,521 Bytes
f83b25e
28bf6cd
 
 
 
2378f58
28bf6cd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a739312
2378f58
28bf6cd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2378f58
28bf6cd
 
 
 
 
2378f58
 
 
 
 
28bf6cd
 
 
 
 
 
 
 
 
 
 
 
a739312
28bf6cd
 
 
 
2378f58
28bf6cd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a739312
 
28bf6cd
 
a739312
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
import gradio as gr
import threading
import os
import pickle
from flask import Flask, request, jsonify
from environment import DocumentClassificationEnv
from baseline_inference import run_task, load_or_train
from agent import TicketAgent

print("Pre-training models...")
models = {}
for d in ["easy", "medium", "hard"]:
    models[d] = load_or_train(d)
    print(f"  {d} model ready!")
print("All models ready!")

agent = TicketAgent(model_dir=".")
print("Agent ready!")

CATEGORIES = {
    "easy":   ["General","Billing","Support","Technical","HR"],
    "medium": ["General","Billing","Support","Technical","HR","Legal","Sales","Marketing","Operations","Complaints"],
    "hard":   [f"Cat_{i}" for i in range(22)]
}

api = Flask(__name__)
api_envs = {}

@api.route("/api/reset", methods=["POST"])
def api_reset():
    data = request.json or {}
    difficulty = data.get("difficulty", "easy")
    seed = data.get("seed", 42)
    env = DocumentClassificationEnv(task_difficulty=difficulty, seed=seed)
    obs, _ = env.reset(seed=seed)
    api_envs["current"] = env
    return jsonify({"observation": {k: v.tolist() if hasattr(v, 'tolist') else v for k, v in obs.items()}})

@api.route("/api/step", methods=["POST"])
def api_step():
    data = request.json or {}
    action = data.get("action", 0)
    env = api_envs.get("current")
    if env is None:
        return jsonify({"error": "Call /api/reset first"}), 400
    obs, reward, done, _, info = env.step(int(action))
    return jsonify({
        "observation": {k: v.tolist() if hasattr(v, 'tolist') else v for k, v in obs.items()},
        "reward": reward, "done": done, "info": info
    })

@api.route("/api/run", methods=["POST"])
def api_run():
    data = request.json or {}
    difficulty = data.get("difficulty", "easy")
    score, t = run_task(difficulty)
    return jsonify({"difficulty": difficulty, "score": score, "time": t})

@api.route("/api/agent/process", methods=["POST"])
def api_agent_process():
    data = request.json or {}
    text = data.get("text", "")
    difficulty = data.get("difficulty", "easy")
    ticket_id = data.get("ticket_id", None)
    if not text:
        return jsonify({"error": "text field required"}), 400
    result = agent.process_ticket(text, difficulty=difficulty, ticket_id=ticket_id)
    return jsonify(result)

@api.route("/api/agent/classify", methods=["POST"])
def api_agent_classify():
    data = request.json or {}
    text = data.get("text", "")
    difficulty = data.get("difficulty", "easy")
    category, scores = agent.classify(text, difficulty)
    priority = agent.get_priority(text)
    return jsonify({"category": category, "priority": priority, "scores": scores})

def run_flask():
    api.run(host="0.0.0.0", port=7861, debug=False)

env_state = {"env": None, "difficulty": "easy"}

def create_env(difficulty):
    e = DocumentClassificationEnv(task_difficulty=difficulty, seed=42)
    obs, _ = e.reset()
    env_state["env"] = e
    env_state["difficulty"] = difficulty
    model = models[difficulty]
    cats = CATEGORIES[difficulty]
    proba = model.predict_proba([obs["content"]])[0]
    top3 = sorted(enumerate(proba), key=lambda x: -x[1])[:3]
    explain = "\n".join([f"  {cats[i] if i < len(cats) else f'Cat_{i}'}: {p*100:.1f}%" for i,p in top3])
    pred_idx = int(model.predict([obs["content"]])[0])
    pred_cat = cats[pred_idx] if pred_idx < len(cats) else f"Cat_{pred_idx}"
    return (f"โœ… Environment created! Difficulty: {difficulty}",
            obs["content"],
            f"๐Ÿค– Model predicts: **{pred_cat}**\n\nTop 3 predictions:\n{explain}",
            "")

def classify_doc(category_idx):
    e = env_state["env"]
    if e is None:
        return "Create environment first!", "", "", ""
    obs, reward, done, _, info = e.step(int(category_idx))
    result = f"Reward: {reward:.2f} | Correct: {info.get('is_correct', False)}"
    if done:
        acc = info.get('episode_accuracy', 0)
        result += f"\n๐Ÿ Episode done! Final Accuracy: {acc:.2%}"
        return result, "Episode complete โ€” create new environment", "", str(info)
    difficulty = env_state["difficulty"]
    model = models[difficulty]
    cats = CATEGORIES[difficulty]
    proba = model.predict_proba([obs["content"]])[0]
    top3 = sorted(enumerate(proba), key=lambda x: -x[1])[:3]
    explain = "\n".join([f"  {cats[i] if i < len(cats) else f'Cat_{i}'}: {p*100:.1f}%" for i,p in top3])
    pred_idx = int(model.predict([obs["content"]])[0])
    pred_cat = cats[pred_idx] if pred_idx < len(cats) else f"Cat_{pred_idx}"
    return (result, obs["content"],
            f"๐Ÿค– Model predicts: **{pred_cat}**\n\nTop 3 predictions:\n{explain}",
            str(info))

def run_baseline_all():
    rows = []
    for d in ["easy", "medium", "hard"]:
        score, t = run_task(d)
        rows.append([d.upper(), f"{score:.4f}", f"{score*100:.1f}%", f"{t:.1f}s"])
    return rows

def process_ticket_ui(ticket_text, difficulty):
    if not ticket_text.strip():
        return "โŒ Please enter ticket text!", "", "", "", "", ""
    result = agent.process_ticket(ticket_text, difficulty=difficulty)
    category_out = f"๐Ÿท๏ธ {result['category']} (confidence: {result['confidence']*100:.1f}%)"
    priority_out  = f"๐Ÿšจ {result['priority'].upper()}"
    dept_out      = f"๐Ÿข {result['department']}\n๐Ÿ“ง {result['email']}\nโฑ๏ธ SLA: {result['sla']}"
    top3_out      = "\n".join([f"  {c}: {p*100:.1f}%" for c,p in result['top3']])
    ref_out       = f"๐ŸŽซ {result['ref_id']} | โฐ {result['timestamp'][:19]}"
    reply_out     = result['reply']
    return category_out, priority_out, dept_out, top3_out, ref_out, reply_out

SAMPLE_TICKETS = {
    "Billing complaint": "My invoice shows an incorrect amount. I was charged $150 but should have been charged $75. Please review and correct this immediately.",
    "Bug report": "The application crashes whenever I try to upload a file larger than 10MB. This is a critical issue affecting my workflow.",
    "HR query": "I have a question about the company's maternity leave policy. How many weeks of paid leave am I entitled to?",
    "Chat message": "Hey, I can't log into my account. The password reset link isn't working either. Please help ASAP!",
    "Email ticket": "Dear Support, I would like to inquire about upgrading my current subscription plan to include additional users.",
}

def load_sample(sample_name):
    return SAMPLE_TICKETS.get(sample_name, "")

with gr.Blocks(title="Document Classification OpenEnv") as demo:
    gr.Markdown("# ๐Ÿ“„ Document Classification OpenEnv")
    gr.Markdown("Real-world customer support ticket routing environment for RL agent training.")

    with gr.Tab("๐Ÿค– Agent Demo"):
        gr.Markdown("### Full Automation โ€” ML Classification + Rule-based Priority + Auto-Reply")
        with gr.Row():
            with gr.Column(scale=2):
                sample_dd = gr.Dropdown(
                    choices=list(SAMPLE_TICKETS.keys()),
                    label="๐Ÿ“‹ Load Sample Ticket",
                    value=None
                )
                ticket_input = gr.Textbox(
                    label="โœ๏ธ Ticket / Message Text",
                    lines=6,
                    placeholder="Paste email, chat message, bug report, HR query..."
                )
                diff_agent = gr.Radio(["easy","medium","hard"], value="medium", label="Model Difficulty")
                btn_process = gr.Button("๐Ÿš€ Process Ticket", variant="primary")
            with gr.Column(scale=2):
                ref_out_box = gr.Textbox(label="๐ŸŽซ Reference ID & Timestamp")
                cat_out_box = gr.Textbox(label="๐Ÿท๏ธ Category & Confidence")
                pri_out_box = gr.Textbox(label="๐Ÿšจ Priority")
                dept_out_box = gr.Textbox(label="๐Ÿข Routed To", lines=3)
                top3_out_box = gr.Textbox(label="๐Ÿ“Š Top 3 Predictions", lines=3)
        reply_out_box = gr.Textbox(label="๐Ÿ“ง Auto-Generated Reply", lines=12)

        sample_dd.change(load_sample, inputs=sample_dd, outputs=ticket_input)
        btn_process.click(
            process_ticket_ui,
            inputs=[ticket_input, diff_agent],
            outputs=[cat_out_box, pri_out_box, dept_out_box, top3_out_box, ref_out_box, reply_out_box]
        )

    with gr.Tab("๐ŸŽฎ Interactive Demo"):
        diff = gr.Radio(["easy","medium","hard"], value="easy", label="Difficulty")
        btn_create = gr.Button("Create Environment", variant="primary")
        status = gr.Textbox(label="Status")
        with gr.Row():
            doc_content = gr.Textbox(label="๐Ÿ“ Document Content", lines=6)
            explain_box = gr.Textbox(label="๐Ÿง  Explainability โ€” Model Reasoning", lines=6)
        category = gr.Number(label="Category Index (easy: 0-4, medium: 0-9, hard: 0-21)", value=0)
        btn_classify = gr.Button("Classify Document")
        result = gr.Textbox(label="Result")
        info_box = gr.Textbox(label="Info")
        btn_create.click(create_env, inputs=diff, outputs=[status, doc_content, explain_box, result])
        btn_classify.click(classify_doc, inputs=category, outputs=[result, doc_content, explain_box, info_box])

    with gr.Tab("๐Ÿ“Š Baseline Evaluation"):
        gr.Markdown("TF-IDF + Logistic Regression baseline โ€” models pre-trained at startup for instant results.")
        btn_eval = gr.Button("Run All Tasks", variant="primary")
        score_table = gr.Dataframe(headers=["Task","Score","Accuracy","Time"], label="Results")
        btn_eval.click(run_baseline_all, outputs=score_table)

    with gr.Tab("๐Ÿ“‹ Environment Info"):
        gr.Markdown("""
## Environment Design
**Task**: Classify customer support documents into correct departments.

| Difficulty | Categories | Episodes | Reward |
|------------|-----------|----------|--------|
| Easy | 5 | 20 | +1.0 correct, -0.4 wrong |
| Medium | 10 | 30 | +1.0 correct, -0.4 wrong |
| Hard | 22 | 50 | +1.0 correct, -0.4 wrong |

## Agent Pipeline
`Ticket Input` โ†’ `ML Classify` โ†’ `Rule Priority` โ†’ `Department Route` โ†’ `Auto Reply`

## API Endpoints (port 7861)
- `POST /api/reset` โ€” `{"difficulty": "easy", "seed": 42}`
- `POST /api/step` โ€” `{"action": 0}`
- `POST /api/agent/process` โ€” `{"text": "my invoice is wrong", "difficulty": "medium"}`
- `POST /api/agent/classify` โ€” `{"text": "urgent bug", "difficulty": "easy"}`
        """)

if __name__ == "__main__":
    t = threading.Thread(target=run_flask, daemon=True)
    t.start()
    demo.launch(server_name="0.0.0.0", server_port=7860)