使用PyCaret做时间序列预测时compare_models函数卡顿问题排查

在Google Colab中使用PyCaret 3.1.0工具，基于汽车经销商零件收入数据集进行时间序列预测时，调用compare_models函数对比模型以寻找最优模型，代码卡在20%进度无法继续（截图显示进度停留在20%）。相关代码如下：

# Only enable critical logging (Optional)
import os
os.environ["PYCARET_CUSTOM_LOGGING_LEVEL"] = "CRITICAL"

def what_is_installed():
    from pycaret import show_versions
    show_versions()

try:
    what_is_installed()
except ModuleNotFoundError:
   !pip install pycaret
   what_is_installed()

import pandas as pd
import numpy as np
import pycaret
pycaret.__version__ # 3.1.0

df = pd.read_csv('parts_revenue.csv', delimiter=';')

from pycaret.utils.time_series import clean_time_index

cleaned = clean_time_index(data=df,
                           index_col='Posting Date',
                           freq='D')

# Verify the resulting DataFrame
print(cleaned.head(n=50))

# parts['MA12'] = parts['Parts Revenue'].rolling(12).mean()


# import plotly.express as px
# fig = px.line(parts, x="Posting Date", y=["Parts Revenue", 
#                "MA12"], template = 'plotly_dark')
# fig.show()

import time
import numpy as np

from pycaret.time_series import *

# We want to forecast the next 12 days of data and we will use 3 
# fold cross-validation to test the models.
fh = 12 # or alternately fh = np.arange(1,13)
fold = 3

# Global Figure Settings for notebook ----
# Depending on whether you are using jupyter notebook, jupyter lab, 
# Google Colab, you may have to set the renderer appropriately
# NOTE: Setting to a static renderer here so that the notebook 
# saved size is reduced.
fig_kwargs = {
              # "renderer": "notebook",
              "renderer": "png",
              "width": 1000,
              "height": 600,
             }

"""## EDA"""

eda = TSForecastingExperiment()
eda.setup(cleaned,
          fh=fh,
          numeric_imputation_target = 0,
          fig_kwargs=fig_kwargs
        )

eda.plot_model()


eda.plot_model(plot="diagnostics",
               fig_kwargs={"height": 800, "width": 1000}
              )

eda.plot_model(
               plot="diff",
               data_kwargs={"lags_list": [[1], [1, 7]],
               "acf": True,
               "pacf": True,
               "periodogram": True},
               fig_kwargs={"height": 800, "width": 1500} )


"""## Modeling"""

exp = TSForecastingExperiment()
exp.setup(data = cleaned,
          fh=fh,
          numeric_imputation_target = 0.0,
          fig_kwargs=fig_kwargs,
          seasonal_period = 5
      )

# compare baseline models
best = exp_ts.compare_models(errors = 'raise') # CODE HANGS HERE!

# plot forecast for 36 months in future
plot_model(best,
           plot = 'forecast',
           data_kwargs = {'fh' : 24}
       )

可能的原因及解决办法 #

• 变量名错误：建模部分初始化了exp = TSForecastingExperiment()，但调用对比模型时使用了未定义的exp_ts，应改为exp.compare_models，这是直接导致代码异常的核心原因。
• 重复实验初始化：先后初始化eda和exp两个实验对象并分别执行setup，可能造成资源占用或环境冲突，建议复用同一个实验对象，或在新实验前清理环境。
• 模型计算负载过高：compare_models默认会训练大量时间序列模型，Colab免费算力下，复杂模型（如深度学习模型）训练耗时极长，看似卡住实则在后台运行。可通过include参数指定轻量模型，比如：

  best = exp.compare_models(include=['arima', 'ets', 'fbprophet'], errors='raise')
  

• 参数与数据不匹配：设置seasonal_period=5但数据是日度频率，若实际业务周期为周（7天）或月，参数不匹配会导致模型训练异常；另外numeric_imputation_target=0的填充逻辑是否合理？若为缺失值，插值填充可能更符合业务逻辑，不当的预处理也会影响模型训练。

内容的提问来源于stack exchange，提问作者AndCh