app.py

import gradio as gr
import pandas as pd
from datetime import datetime

from sentiment import analyze_sentiment
from reply_style import warm_reply


def sentiment_to_numeric(label: str, score: float) -> float:
    label = label.lower()  
    if label == "positive":
        return round(score, 3)
    elif label == "negative":
        return round(-score, 3)
    else:  # neutral
        return 0.0


def confidence_level(score: float) -> str:
    score = float(score)
    if score < 0.4:
        return "low"
    elif score < 0.7:
        return "medium"
    return "high"


def chatbot_response(user_input, chat_messages, mood_log, memory):
    # State 初始化（避免 None）
    chat_messages = chat_messages or []   # list[dict]
    mood_log = mood_log or []             # list[dict]
    memory = memory or []                 # list[dict] (Groq messages)

    user_input = (user_input or "").strip()
    if not user_input:
        chat_messages = chat_messages + [{"role": "assistant", "content": "（你沒有輸入內容哦）"}]
        return chat_messages, chat_messages, mood_log, memory, ""

    # 1) 顯示：使用者訊息
    chat_messages = chat_messages + [{"role": "user", "content": user_input}]

    # 2) 情緒分析
    sentiment, score = analyze_sentiment(user_input)

    # 3) 生成回覆（warm_reply 必須回傳 reply, memory）
    reply, memory = warm_reply(user_input, sentiment, score, memory)

    # 4) 顯示：助理回覆
    chat_messages = chat_messages + [{"role": "assistant", "content": reply}]

    # 5) 記錄心情（含轉換後欄位）
    sentiment_numeric = sentiment_to_numeric(sentiment, score)
    confidence_lvl = confidence_level(score)

    if mood_log:
        prev = mood_log[-1]
        delta_sentiment = sentiment_numeric - float(prev.get("sentiment_numeric", 0.0))
        delta_confidence = float(score) - float(prev.get("confidence_score", 0.0))
    else:
        delta_sentiment = 0.0
        delta_confidence = 0.0

    mood_log.append({
        "timestamp": datetime.now().strftime("%Y-%m-%d %H:%M:%S"),
        "text": user_input,
        "sentiment_label": sentiment,
        "sentiment_numeric": sentiment_numeric,
        "confidence_score": round(float(score), 3),
        "confidence_level": confidence_lvl,
        "delta_sentiment": round(delta_sentiment, 3),
        "delta_confidence": round(delta_confidence, 3),
    })

    # 最後回傳 msg="" 清空輸入框
    return chat_messages, chat_messages, mood_log, memory, ""


def export_csv(mood_log):
    if not mood_log:
        return None

    df = pd.DataFrame(mood_log)
    columns = [
        "timestamp",
        "text",
        "sentiment_label",
        "sentiment_numeric",
        "confidence_score",
        "confidence_level",
        "delta_sentiment",
        "delta_confidence",
    ]
    for c in columns:
        if c not in df.columns:
            df[c] = None
    df = df[columns]

    df.to_csv("mood_analysis.csv", index=False, encoding="utf-8-sig")
    return "mood_analysis.csv"


def reset_session():
    # 回傳：chat_state, mood_state, memory_state, msg, chatbot
    return [], [], [], "", []


with gr.Blocks() as demo:
    gr.Markdown("# 跟我聊聊天")

    chatbot = gr.Chatbot(height=420)  

    chat_state = gr.State([])     # messages list (dict)
    mood_state = gr.State([])     # mood log list (dict)
    memory_state = gr.State([])   # groq memory list (dict)

    with gr.Row():
        msg = gr.Textbox(placeholder="輸入訊息…", show_label=False, scale=5)
        submit = gr.Button("送出", scale=1)

    msg.submit(
        chatbot_response,
        inputs=[msg, chat_state, mood_state, memory_state],
        outputs=[chatbot, chat_state, mood_state, memory_state, msg],
    )
    submit.click(
        chatbot_response,
        inputs=[msg, chat_state, mood_state, memory_state],
        outputs=[chatbot, chat_state, mood_state, memory_state, msg],
    )

    with gr.Row():
        export_btn = gr.Button("匯出心情記錄 CSV")
        download_file = gr.File(label="File")

    export_btn.click(export_csv, inputs=[mood_state], outputs=[download_file])

    with gr.Row():
        reset_btn = gr.Button("清除本次對話/記憶")

    reset_btn.click(
        reset_session,
        inputs=None,
        outputs=[chat_state, mood_state, memory_state, msg, chatbot],
    )

demo.launch()