如何借助Streamlit Session State实现多页面共享上传Excel文件

核心思路 #

用Streamlit内置的st.session_state跨页面共享上传后的DataFrame，主页面上传文件后将原始数据存入session_state，其他页面直接读取该共享数据，无需重复上传。

修改步骤 #

• 主页面上传文件后，把原始DataFrame存入st.session_state（避免后续页面修改数据影响其他页面）
• 其他页面先检查session_state中是否存在共享数据，存在则直接使用，不存在则提示用户前往主页面上传
• 移除其他页面的文件上传组件，或保留组件但提示无需重复操作

修改后的代码示例 #

主页面代码 #

import streamlit as st
import pandas as pd
import numpy as np

# 重置功能：清空session_state并刷新页面
if st.sidebar.button("Reset Page"):
    for key in st.session_state.keys():
        del st.session_state[key]
    st.experimental_rerun()

st.image('https://wallpapercave.com/wp/9WBdkYu.jpg')
st.title('Excel Drag and Drop')
uploaded_file = st.file_uploader('Please upload your Excel File')

if uploaded_file:
    # 读取原始数据并存入session_state
    raw_df = pd.read_excel(uploaded_file)
    st.session_state['uploaded_df'] = raw_df.copy()

    # 主页面System Three处理逻辑
    df = raw_df.copy()
    st.subheader('System Three Selections')

    df.drop(df[df["Country"] == "South Africa"].index, inplace=True)
    df = df.drop('Date', axis=1)

    df.replace('-', 0, inplace=True)
    df["Win Result"] = df["Win Result"].replace(["L"], -1)
    df["True Odds"] = np.where(df["BSP"].ne(0), df["BSP"].sub(1), 0)
    df["ProfitLoss"] = np.where(df["Win Result"].eq(-1), df["Win Result"], df["True Odds"])

    df.replace('-', 0, inplace=True)
    df.loc[df['Win Result'] ==  0, 'True Odds'] = 0
    df.replace('-', 0, inplace=True)

    df.loc[df['BSP'] == 0, 'ProfitLoss'] = 0
    df.loc[df['Win Result'] == 0, 'ProfitLoss'] = 0
    df = df[(df['Country'] == 'United Kingdom') &
            (df['WFF Rank'] == 1) &
            (df['OR'] >= 100) &
            (df['BSP'].between(1.5, 30))
    ]
    df.loc["Total", "ProfitLoss"] = df.ProfitLoss.sum()

    st.write(df)

Page1代码 #

import streamlit as st
import pandas as pd
import numpy as np

st.set_page_config(layout="wide")

# 重置功能：清空session_state并刷新页面
if st.sidebar.button("Reset Page"):
    for key in st.session_state.keys():
        del st.session_state[key]
    st.experimental_rerun()

st.image('https://wallpapercave.com/wp/9WBdkYu.jpg')
st.title('Excel Drag and Drop')

# 检查是否有共享数据
if 'uploaded_df' in st.session_state:
    # 读取共享的原始数据
    raw_df = st.session_state['uploaded_df']
    st.success("已使用主页面上传的Excel文件")

    # Page1的System Two处理逻辑
    df = raw_df.copy()
    st.subheader('System Two Selections')
    df['Last Form'] = pd.to_numeric(df['Last Form'].replace({'U': '', 'C': '', 'B':'', 'D': '',
                                                             'F': '', '-': 0, 'P': '', 'R': '',
                                                             'S': '', 'V': ''}), errors='coerce')
    df.drop(df[df["Country"] == "South Africa"].index, inplace=True)
    df = df[~df['Venue'].isin(['Kempton', 'Kilbeggan', 'Kilarney', 'Leicester', 'Leopardstown', 'Limerick',
                               'Aintree'])]
    df = df[~df['Race Distance'].isin(['1m 6f 17', '1m 6y'])]

    df.replace('-', 0, inplace=True)
    df["Win Result"] = df["Win Result"].replace(["L"], -1)
    df["True Odds"] = np.where(df["BSP"].ne(0), df["BSP"].sub(1), 0)
    df["ProfitLoss"] = np.where(df["Win Result"].eq(-1), df["Win Result"], df["True Odds"])
    df.loc[df['Win Result'] == 0, 'True Odds'] = 0
    df.replace('-', 0, inplace=True)

    df.loc[df['BSP'] == 0, 'ProfitLoss'] = 0
    df.loc[df['Win Result'] == 0, 'ProfitLoss'] = 0
    df = df[(df['Country'] == 'United Kingdom') &
            (df['WFF Rank'] == 1) &
            (df['Last Form'].between(0, 3, inclusive=False)) &
            (df['Class Change'] >= 0) &
            (df['Improving Horse'] == 'Y') &
            (df['RDB Rating'] >= 140) &
            (df['Age'] <= 7) &
            (df['Runner Count'] <= 20) &
            (df['RDB Rank'] == 1)
            ]
    df.loc["Total", "ProfitLoss"] = df.ProfitLoss.sum()

    st.write(df)
    st.download_button(label='Download System 2', data=df.to_csv(), mime='text/csv')
else:
    st.warning("请先到主页面上传Excel文件")

关键说明 #

• 存储原始DataFrame到session_state，每个页面基于副本做独立处理，避免页面间数据互相干扰
• 重置按钮改用Streamlit原生的st.experimental_rerun()，比pyautogui刷新更符合框架交互逻辑
• Page2至Page5可完全参照Page1的逻辑修改，实现统一的共享数据读取逻辑

内容的提问来源于stack exchange，提问作者liambh