使用scikit-learn/XGBoost作为MICE模型时fancyimpute报TypeError求助

Hey there, I’ve run into this exact issue before! The problem here is how you’re passing the model to MICE — you’re giving it an already-instantiated object (like RandomForestRegressor()), but fancyimpute expects you to pass the model class itself instead.

Why this happens #

Under the hood, MICE creates multiple instances of your model (one for each feature with missing values) and manages initialization parameters internally. When you pass a pre-made instance, it tries to reuse it in ways that break the expected interface, leading directly to that TypeError.

The Fix #

Instead of passing RandomForestRegressor() (with parentheses), pass just RandomForestRegressor (the class reference). The same rule applies to XGBoost’s XGBRegressor — pass the class, not an instance.

Here’s your corrected code:

from fancyimpute import MICE
from sklearn.ensemble import RandomForestRegressor
import pandas as pd

# Pass the model CLASS, not an instantiated object
mice = MICE(n_imputations=1, verbose=False, model=RandomForestRegressor)
df = pd.DataFrame(data=mice.complete(d), columns=d.columns, index=d.index)

For XGBoost Users #

If you want to use XGBoost instead, the approach is identical — make sure to pass the XGBRegressor class (from the xgboost library) without instantiating it first:

from xgboost import XGBRegressor

mice = MICE(n_imputations=1, verbose=False, model=XGBRegressor)
df = pd.DataFrame(data=mice.complete(d), columns=d.columns, index=d.index)

Bonus: Customizing Model Parameters #

If you want to set specific hyperparameters for your model (like n_estimators for RandomForest), use the model_kwargs parameter in MICE to pass them in:

mice = MICE(
    n_imputations=1,
    verbose=False,
    model=RandomForestRegressor,
    model_kwargs={"n_estimators": 200, "random_state": 42}
)

That should resolve the TypeError you’re seeing. Let me know if you hit any other snags!

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