Update app.py

app.py

@@ -3,60 +3,43 @@ import gradio as gr
 import xgboost as xgb
 import numpy as np
 import pandas as pd
-import requests
+from transformers import pipeline
 from huggingface_hub import hf_hub_download
 
-# **📌 LLM 服务器地址（修改为你的 LLM Space 地址）**
-LLM_API_URL = "https://your-llm-space.gradio.app"  # 替换为 LLM 服务器地址
+# **📌 先运行 `preprocess.py`，确保数据可用**
+processed_data_path = "processed_data.csv"
 
-# **📌 先运行 `preprocess.py` 处理数据**
-if not os.path.exists("processed_data.csv"):
+if not os.path.exists(processed_data_path):
     print("📌 运行 `preprocess.py` 进行数据处理...")
     os.system("python preprocess.py")
 
 # **📌 加载处理后的数据**
-processed_data_path = "processed_data.csv"
-if not os.path.exists(processed_data_path):
-    raise FileNotFoundError("❌ `processed_data.csv` 未找到，请先运行 `preprocess.py` 处理数据！")
-
 df = pd.read_csv(processed_data_path)
 
+# **📌 加载 Hugging Face GPT-Neo 作为 LLM**
+print("📌 正在加载 GPT-Neo 模型...")
+generator = pipeline('text-generation', model='EleutherAI/gpt-neo-2.7B')
+print("✅ GPT-Neo 模型加载成功！")
+
 # **📌 加载 XGBoost 预测模型**
 model_path = hf_hub_download(repo_id="YDluffy/lottery_prediction", filename="lottery_xgboost_model.json")
 model = xgb.XGBRegressor()
 model.load_model(model_path)
 
-# **📌 调用 LLM API 解析用户输入**
-def get_prediction_from_llm(user_input):
+# **📌 LLM 解析用户输入**
+def chat_with_llm(user_input):
+    prompt = f"请从以下问题中提取出预测所需的参数（年份、期号、中奖号码等）：'{user_input}'"
+    
     try:
-        # 发送请求到 LLM API
-        response = requests.post(LLM_API_URL + "/api/predict", json={"input": user_input})
-
-        # **确保返回 JSON**
-        if response.status_code == 200:
-            try:
-                llm_output = response.json().get("prediction", "")
-            except requests.exceptions.JSONDecodeError:
-                return "❌ LLM 服务器返回了无效的 JSON 数据，请检查 LLM Space 是否正常运行。"
-        else:
-            return f"❌ LLM 服务器错误，状态码: {response.status_code}"
-
-    except requests.exceptions.ConnectionError:
-        return "❌ 无法连接到 LLM 服务器，请确保 LLM Space 正在运行。"
-
-    # **📌 解析 LLM 输出**
-    year, period = 2025, 16
-    nums = [5, 12, 23, 34, 45, 56]
-    special = 7
-
-    # **📌 进行 XGBoost 预测**
-    prediction = predict_lottery(year, period, *nums, special)
-
-    return f"📊 预测的号码是: {prediction}\n\n📢 LLM 解析的特征：{llm_output}"
+        response = generator(prompt, max_length=100, num_return_sequences=1)
+        extracted_text = response[0]['generated_text']
+        return extracted_text
+    except Exception as e:
+        return f"❌ GPT-Neo 处理错误: {str(e)}"
 
 # **📌 预测函数**
 def predict_lottery(year, period, num1, num2, num3, num4, num5, num6, special):
-    # 从历史数据查找对应的月份和日期
+    # **从历史数据查找对应的月份和日期**
     history = df[(df['期号_年份'] == year) & (df['期数'] == period)]
     if not history.empty:
         month = history.iloc[0]['月份']
@@ -64,19 +47,33 @@ def predict_lottery(year, period, num1, num2, num3, num4, num5, num6, special):
     else:
         month, day = 1, 1  # 默认值
 
-    # 计算中奖号码均值
+    # **计算中奖号码均值**
     avg_number = np.mean([num1, num2, num3, num4, num5, num6])
 
-    # 形成特征数组
+    # **形成特征数组**
     features = np.array([[year, period, month, day, num1, num2, num3, num4, num5, num6, special, avg_number]])
 
-    # 进行预测
+    # **进行预测**
     prediction = model.predict(features)
     return prediction
 
+# **📌 结合 LLM 和 XGBoost**
+def predict_and_interact(user_input):
+    llm_output = chat_with_llm(user_input)
+
+    # **默认解析出的预测参数**
+    year, period = 2025, 16
+    nums = [5, 12, 23, 34, 45, 56]
+    special = 7
+
+    # **进行预测**
+    prediction = predict_lottery(year, period, *nums, special)
+
+    return f"📊 预测的号码是: {prediction}\n\n📢 LLM 解析的特征：{llm_output}"
+
 # **📌 Gradio Web 界面**
 iface = gr.Interface(
-    fn=get_prediction_from_llm,
+    fn=predict_and_interact,
     inputs=gr.Textbox(label="请输入问题或期号信息"),
     outputs="text",
     title="六合彩预测模型",