ARIMA预测报错：无法将float64转为int64（same_kind规则）

Hey, let's work through this error you're facing! That Cannot cast ufunc subtract output from dtype('float64') to dtype('int64') with casting rule 'same_kind' message boils down to a data type mismatch between your input data and the ARIMA model's internal calculations.

What's causing this? #

Your specific dataset is stored as int64 (integer) values, while other datasets you're using are likely floating-point. When you save the ARIMA model and reload it to run forecasts, the model tries to perform subtraction operations that produce float64 results—but it's attempting to store those results in a structure expecting integers. Python's same_kind casting rule blocks this implicit conversion (to avoid accidental data loss), hence the error.

The fix is simple #

All you need to do is convert your input data to floating-point before fitting the model. Here's your revised code with the key fix plus a minor syntax correction:

import pandas as pd
from statsmodels.tsa.arima.model import ARIMA
from statsmodels.tsa.arima.model import ARIMAResults

def forecast_fit(df):
    series = df
    # Critical fix: Convert integer data to float64 to match model's internal math
    X = series.values.astype('float64')
    train = X
    model = ARIMA(X, order=(1,0,1))
    model_fit = model.fit(disp=0)
    model_fit.save('model.pkl')

# Fit the model with your target dataframe
forecast_fit(df)

# Out-of-sample forecast (fixed the missing closing quote here too!)
loaded = ARIMAResults.load('model.pkl')
forecast = loaded.forecast(steps=17)[0]
df_forecast = pd.DataFrame(forecast, columns=[f"{col}_hat" for col in df.columns])

Quick additional notes #

• I fixed a tiny syntax error in your original code: loaded = ARIMAResults.load('model.pkl) was missing a closing single quote—easy typo to overlook!
• If you're using an older version of statsmodels (pre-0.12), adjust the imports to from statsmodels.tsa.arima_model import ARIMA, ARIMAResults instead.
• Converting integers to float64 doesn't lose any data, and it ensures the model's floating-point operations don't hit type conflicts during forecasting.

内容的提问来源于stack exchange，提问作者Ankit Tripathi