s5.py

# -*- coding: utf-8 -*-
"""
Created on Mon Apr 24 15:41:55 2023

@author: 24397
"""

import streamlit as st
import pandas as pd
import numpy as np
from sklearn.neural_network import MLPRegressor
from sklearn.preprocessing import StandardScaler
from sklearn.model_selection import train_test_split
import matplotlib.pyplot as plt
from sklearn.metrics import r2_score,mean_squared_error
import seaborn as sns
from sklearn.neural_network import MLPRegressor
from sklearn.utils.validation import check_X_y, check_array, check_is_fitted
plt.rcParams['font.family'] = 'SimHei' # 设置字体为SimHei

# 上传文件并选择要使用的列数
uploaded_file = st.file_uploader("上传Excel文件", type="xlsx")
if uploaded_file is not None:
    data = pd.read_excel(uploaded_file)
    input_cols = st.multiselect('请选择输入列', list(data.columns), default=list(data.columns)[:2])
    output_cols = st.multiselect('请选择输出列', list(data.columns), default=list(data.columns)[-1:])
    X = data[input_cols].values
    y = data[output_cols].values
# 数据预处理
    scaler = StandardScaler()
    X_train_raw, X_test_raw, y_train_raw, y_test_raw = train_test_split(data[input_cols].values,
                                                                    data[output_cols].values,
                                                                            test_size=0.2,
                                                                            random_state=None)
    X_train_scaled = scaler.fit_transform(X_train_raw)
    X_test_scaled = scaler.transform(X_test_raw)
    # 调整参数
    hidden_layers = st.slider('隐藏层层数', min_value=1, max_value=10, value=3)
    nodes_list = []
    for i in range(hidden_layers):
        nodes_list.append(st.number_input(f'隐藏层{i+1}节点数', min_value=1, max_value=1000))
    activation = st.selectbox('激励函数', options=['relu', 'logistic', 'tanh','identity'], index=0)
    solver = st.selectbox('优化函数', options=['lbfgs', 'sgd', 'adam'], index=2)
    alpha = st.number_input('正则化系数', value=0.0001)
    learning_rate_init = st.number_input('学习率', value=0.001)
    iterations = st.number_input('迭代次数', value=100)


class CustomMLPRegressor(MLPRegressor):
    def fit(self, X, y):
        X, y = check_X_y(X, y, accept_sparse=['csr', 'csc', 'coo'])
        self._label_binarizer = None
        self._random_state = np.random.RandomState(self.random_state)
        
        if self.verbose:
            print(f"Training loss at epoch 1: {self.loss_}")

        self.losses_ = []
        
        def record_loss(*args, **kwargs):
            self.losses_.append(self.loss_)
        
        self._stopping_criterion_callback = record_loss
        
        super().fit(X, y)
        return self
    
    def _stopping_criterion_callback(self, *args, **kwargs):
        pass



model = CustomMLPRegressor(hidden_layer_sizes=tuple(nodes_list), activation=activation, solver=solver,
                          alpha=alpha, learning_rate_init=learning_rate_init, max_iter=iterations)
 # Replace with your data
model.fit(X, y)

# Save losses to Excel
df1 = pd.DataFrame(model.losses_, columns=['Loss'])
df1.to_excel('training_losses.xlsx', index=False)
save_button = st.button("保存Loss")





   # 训练模型并进行预测
model.fit(X_train_scaled, y_train_raw)
y_pred_raw = model.predict(X_test_scaled)
r2score = r2_score(y_test_raw,y_pred_raw)
mse = mean_squared_error(y_test_raw,y_pred_raw)

# 显示模型评估指标的结果
st.write("## 模型评估")
st.write(f"R2 score: {r2score:.4f}")
st.write(f"均方误差(MSE): {mse:.4f}")
st.write("y:")
st.write(y_test_raw)
st.write("y_pred:")
st.write(y_pred_raw)

# 添加保存按钮
save_button = st.button("保存数据")

if save_button:
    # 将 y 和 y_pred 组合成 DataFrame
    df = pd.DataFrame({'y': y_test_raw.flatten(), 'y_pred': y_pred_raw.flatten()})
    file_name = st.text_input("请输入文件名")
    if file_name:
        df.to_excel(f"{file_name}.xlsx", index=False)
        st.write(f"数据已保存为 {file_name}.xlsx")
  # 展示预测结果和真实值的对比图像
st.write("真实值与预测值的对比图像：")
fig, axs = plt.subplots(len(output_cols), 1, figsize=(6, 4 * len(output_cols)))
if len(output_cols) == 1:
            axs.plot(y_test_raw, label="真实值")
            axs.plot(y_pred_raw, label="预测值")
            axs.set_title("{}".format(output_cols[0]))
            axs.legend()
else:
            for i in range(len(output_cols)):
                axs[i].plot(y_test_raw[:, i], label="真实值")
                axs[i].plot(y_pred_raw[:, i], label="预测值")
                axs[i].set_title("{}".format(output_cols[i]))
                axs[i].legend()
st.pyplot(fig)