基于TabPFN算法的回归问题-代码运行
一、源码链接
github链接下载到源码:https://github.com/PriorLabs/TabPFN?tab=readme-ov-file
二、示例代码
# -*- coding: utf-8 -*- # TabPFN 回归预测 # Author: HAN import pandas as pd import numpy as np import matplotlib.pyplot as plt from sklearn.model_selection import train_test_split from sklearn.metrics import r2_score, mean_squared_error from tabpfn import TabPFNRegressor # 读取数据 csv_path = r"H:\11_tranfer_model\Local_data\3_tranfer_model\data_china.csv" df = pd.read_csv(csv_path) # 定义变量 target_col = "ECE" feature_cols = [ "SI", "SI1", "SI2", "SI3", "SI4", "S1", "S2", "S3", "S5", "S6", "S7", "S8", "S9", "CRSI", "NDSI", "SI_T", "NDVI", "ENDVI", "DVI", "GDVI", "RVI" ] # 构建数据 X = df[feature_cols] y = df[target_col] # 删除缺失值 data = pd.concat([X, y], axis=1).dropna() X = data[feature_cols].values y = data[target_col].values # 训练测试划分 X_train, X_test, y_train, y_test = train_test_split( X, y, test_size=0.2, random_state=42 ) # 训练 TabPFN model = TabPFNRegressor() model.fit(X_train, y_train) y_pred = model.predict(X_test) # 模型评价 r2 = r2_score(y_test, y_pred) rmse = np.sqrt(mean_squared_error(y_test, y_pred)) print("R2 =", r2) print("RMSE =", rmse) # 绘制预测散点图 plt.figure(figsize=(6,6)) plt.scatter(y_test, y_pred, alpha=0.7) # 1:1 line min_val = min(y_test.min(), y_pred.min()) max_val = max(y_test.max(), y_pred.max()) plt.plot([min_val, max_val],[min_val, max_val], linestyle="--") plt.xlabel("Observed ECe") plt.ylabel("Predicted ECe") plt.title(f"TabPFN Prediction\nR² = {r2:.3f} RMSE = {rmse:.3f}") plt.tight_layout() plt.show()三、代码运行
1、安装tabpfn包
pip install tabpfn==6.2.1
2、Hugging Face 账号授权
2.1 打开：https://huggingface.co/，注册 / 登录账号
2.2 访问模型页面：https://huggingface.co/Prior-Labs/tabpfn_2_5，点击Access repository,接受协议
2.3 访问页面：https://huggingface.co/settings/tokens
账号---Setting -----Access Token -----点击+Create new token-----token name:tabpfn_token-------token type: read ------点击create token----复制 token
3、Hugging Face登录
from huggingface_hub import login, snapshot_download # ========== 关键：替换成刚复制的 Token ========== MY_TOKEN = "hf_*******" # 1. 登录（跳过终端，直接代码验证） try: login(token=MY_TOKEN, add_to_git_credential=False) print("登录成功！") except Exception as e: print(f"登录失败：{e}") exit()显示登录成功后，可运行{二、示例代码}
4、备注
第一次使用或者新环境/没有缓存时，必须使用外网，必须登录 Hugging Face，授权访问模型。登录成功后，会在本地.cache里生成认证信息，之后就不用每次重新登录了（除非换环境、删缓存或者 token 过期），也不需要外网。