PyCaret时间序列在Streamlit中无法同时显示样本内外绘图

在使用PyCaret时间序列模块开发Streamlit应用做数据实验时，遇到样本外（out-of-sample）绘图正常显示，但样本内（insample）绘图无法显示的问题，原代码如下：

from pycaret.time_series import TSForecastingExperiment
import streamlit as st
exp_auto = TSForecastingExperiment()
def run_pycaret(df, target='qty', horizon=12):
    exp_auto.setup(
        data=df, target=target, fh=horizon, enforce_exogenous=False,
        numeric_imputation_target="ffill", numeric_imputation_exogenous="ffill",
        scale_target="minmax",
        scale_exogenous="minmax",
        fold_strategy="expanding",
        session_id=42, verbose=False)
    best = exp_auto.compare_models(sort="mape", turbo=False, verbose=False)
    metrics = exp_auto.pull()
    insample = exp_auto.plot_model(best, plot='insample',display_format='streamlit')
    outsample = exp_auto.plot_model(best, plot='forecast', display_format='streamlit')
    return best, metrics, insample, outsample


if st.button("Run Time series models:"):
    best, metrics, insample, outsample = run_pycaret(df_exog, horizon=steps)

    st.write(metrics)
    # Plot graph
    col1_ts_exp, col2_ts_exp = st.columns(2)
    with col1_ts_exp:
        insample
    with col2_ts_exp:
        outsample

实验结果仅显示样本外绘图：

问题原因 #

PyCaret的plot_model方法在指定display_format='streamlit'时，针对不同绘图类型的处理逻辑存在差异：

• 对于forecast类型的绘图，方法会返回可赋值的绘图对象；
• 对于insample类型的绘图，方法会直接在当前Streamlit上下文渲染，不会返回有效对象，赋值给insample的实际是None，因此无法显示。

解决方案 #

不要将plot_model的结果赋值给变量，而是直接在对应的Streamlit列上下文里调用plot_model方法。修改后的代码如下：

from pycaret.time_series import TSForecastingExperiment
import streamlit as st
exp_auto = TSForecastingExperiment()
def run_pycaret(df, target='qty', horizon=12):
    exp_auto.setup(
        data=df, target=target, fh=horizon, enforce_exogenous=False,
        numeric_imputation_target="ffill", numeric_imputation_exogenous="ffill",
        scale_target="minmax",
        scale_exogenous="minmax",
        fold_strategy="expanding",
        session_id=42, verbose=False)
    best = exp_auto.compare_models(sort="mape", turbo=False, verbose=False)
    metrics = exp_auto.pull()
    # 不再赋值insample和outsample，直接在Streamlit上下文渲染
    return best, metrics


if st.button("Run Time series models:"):
    best, metrics = run_pycaret(df_exog, horizon=steps)

    st.write(metrics)
    # Plot graph
    col1_ts_exp, col2_ts_exp = st.columns(2)
    with col1_ts_exp:
        st.subheader("样本内拟合图")
        exp_auto.plot_model(best, plot='insample', display_format='streamlit')
    with col2_ts_exp:
        st.subheader("样本外预测图")
        exp_auto.plot_model(best, plot='forecast', display_format='streamlit')

修改后，样本内和样本外的绘图就能正常在对应的列中显示。

内容的提问来源于stack exchange，提问作者Sam.H