咨询：用Streamlit/Dash实现多依赖动态下拉筛选的可行性

项目背景与需求 #

正在开发一个小项目，需要用Python可视化DataFrame，要求用户能对continent、country、language三个存在相互依赖关系的列进行筛选。

示例DataFrame代码如下：

import pandas as pd

df = pd.DataFrame(
    {
        "continent": ["Asia"] * 2 + ["Europe"] * 2 + ["North America"] * 2,
        "country": ["China", "Japan"]
        + ["United Kingdom", "France"]
        + ["United States", "Canada"],
        "language": ["Chinese", "Japanese"]
        + ["English", "French"]
        + ["English", "English"],
    }
)

具体筛选需求 #

• 用户选择continent = North America后，country选项应缩小为United States和Canada，language选项仅保留English；若切换为continent = Europe，country和language选项需同步更新为欧洲对应内容
• 用户先选择language = English后，continent选项应缩小为North America和Europe（使用英语的国家为Canada、United States和United Kingdom）

这类筛选存在非线性依赖关系。

Streamlit尝试情况 #

尝试用Streamlit结合callback实现，但发现Streamlit似乎不适合处理复杂依赖场景。

已用Streamlit实现一个MVP，下拉菜单具备动态效果，但由于选项依赖筛选后的DataFrame，用户选择后只能通过“Refresh Data”按钮重置，体验不佳。代码如下：

import pandas as pd
import streamlit as st

st.set_page_config(page_title="Toy App", layout="wide")


@st.cache
def get_data():
    df = pd.DataFrame(
        {
            "Continent": ["Asia"] * 2 + ["Europe"] * 2 + ["North America"] * 2,
            "Country": ["China", "Japan"]
            + ["United Kingdom", "France"]
            + ["United States", "Canada"],
            "Language": ["Chinese", "Japanese"]
            + ["English", "French"]
            + ["English", "English"],
        }
    )
    return df


def update_df(df: pd.DataFrame) -> pd.DataFrame:
    continent = st.session_state["Continent"]
    country = st.session_state["Country"]
    language = st.session_state["Language"]
    if continent != "all":
        df = df.query(f"Continent == '{continent}'")
    if country != "all":
        df = df.query(f"Country == '{country}'")
    if language != "all":
        df = df.query(f"Language == '{language}'")
    st.session_state["df"] = df
    st.session_state["fresh_data"] = False


df = get_data()


if "df" not in st.session_state:
    st.session_state.df = df
if "fresh_data" not in st.session_state:
    st.session_state.fresh_data = True

with st.expander("Display", expanded=True):
    if st.button("Refresh data"):
        df = get_data()
        st.session_state["df"] = df
        st.session_state["fresh_data"] = True

    df = st.session_state["df"]
    col1, col2, col3 = st.columns(3)

    continent_options = df.Continent.unique().tolist()
    country_options = df.Country.unique().tolist()
    language_options = df.Language.unique().tolist()

    if st.session_state.fresh_data:
        country_options.insert(0, "all")
        language_options.insert(0, "all")
        continent_options.insert(0, "all")

    continent = col1.selectbox(
        "Continent",
        options=continent_options,
        on_change=update_df,
        kwargs={"df": df},
        key="Continent",
    )
    countries = col2.selectbox(
        "Country",
        options=country_options,
        on_change=update_df,
        kwargs={"df": df},
        key="Country",
    )
    language = col3.selectbox(
        "Language",
        options=language_options,
        on_change=update_df,
        kwargs={"df": df},
        key="Language",
    )

    st.write(df.astype("object"))

核心疑问 #

是否可以用Dash构建这类复杂依赖的动态下拉筛选？

内容的提问来源于stack exchange，提问作者bayes2021