使用Requests加载GitHub机器学习模型时出现AttributeError问题求助

错误原因 #

• 保存方式错误：requests.get(url).content返回的是字节流，你用pickle.dump把字节流存成文件，后续pickle.load读取到的只是原始字节数据，不是训练好的模型对象，自然没有predict方法。
• 变量名拼写错误：代码里pd.DataFrame(x)中的x未定义（应该是input_array）；modell.predict中的modell是笔误，应为model。

修正后的代码 #

import requests
import pickle
import os
from os.path import getcwd
import numpy as np
import pandas as pd

# 下载模型文件
url = "https://github.com/MUmairAB/Stroke-Prediction-using-Machine-Learning/blob/main/trained_model.sav?raw=true"
r = requests.get(url)
# 直接将字节流写入文件，而非用pickle.dump
with open(os.path.join(getcwd(), 'trained_model12.pkl'), 'wb') as f:
    f.write(r.content)

# 加载模型
model = pickle.load(open(os.path.join(getcwd(), 'trained_model12.pkl'), 'rb'))

# 准备输入数据并预测
input_array = np.array([[0,67,0,1,0,228.69,36.6],
                        [0,58,1,0,0,87.96,39.2]])
input_df = pd.DataFrame(input_array)
prediction = model.predict(input_df)
print(prediction)

其他可行方案 #

方案1：使用urllib下载 #

和requests逻辑一致，换用urllib库实现：

import urllib.request
import pickle
import os
from os.path import getcwd

url = "https://github.com/MUmairAB/Stroke-Prediction-using-Machine-Learning/blob/main/trained_model.sav?raw=true"
urllib.request.urlretrieve(url, os.path.join(getcwd(), 'trained_model12.pkl'))
model = pickle.load(open(os.path.join(getcwd(), 'trained_model12.pkl'), 'rb'))

方案2：在.py文件中执行git clone #

如果习惯用git clone，可通过subprocess模块在Python脚本中执行克隆命令：

import subprocess
import pickle
import os
from os.path import getcwd

repo_url = "https://github.com/MUmairAB/Stroke-Prediction-using-Machine-Learning.git"
clone_path = os.path.join(getcwd(), 'stroke_model_repo')
# 克隆仓库
subprocess.run(["git", "clone", repo_url, clone_path], check=True)
# 加载模型
model_path = os.path.join(clone_path, 'trained_model.sav')
model = pickle.load(open(model_path, 'rb'))

内容的提问来源于stack exchange，提问作者Umair Akram