TPOT:使用TPOTRegressor时出现酸洗错误

Question

我有一个名为X的DataFrame和一组名为Y的目标值。

对于我的大多数模型，我都是这样做的(只是一个例子)：

from sklearn.linear_model import LassoCV
clf = LassoCV()
score = cross_val_score(estimator = clf, X = X, y = Y, cv = KFold(n_splits = 3, random_state = 100), n_jobs = -1, \
                        scoring = "neg_mean_squared_error")
np.mean([np.sqrt(-x) for x in score])

我正在尝试以类似的方式使用TPOT，如下所示：

from tpot import TPOTRegressor
tpot = TPOTRegressor(generations=20, population_size=100, verbosity=2)

score = cross_val_score(estimator = tpot, X = X, y = Y, cv = KFold(n_splits = 3, random_state = 100), n_jobs = -1, \
                        scoring = "neg_mean_squared_error")
np.mean([np.sqrt(-x) for x in score])

TPOT启动，但随后显示以下酸洗错误：

PicklingError: Can't pickle <type 'instancemethod'>: it's not found as __builtin__.instancemethod

你知道为什么会发生这种情况/如何让TPOT更好地发挥作用吗？

谢谢!

Answer 1

如果您使用的是Python 2，请尝试：

import dill  

这样就可以对lambda函数进行酸洗……对我来说很有效。

在Python 3中，您可能需要：

import dill as pickle

Answer 2

尝试使用：tpot.fitted_pipeline_

from tpot import TPOTRegressor
tpot = TPOTRegressor(generations=20, population_size=100, verbosity=2)

score = cross_val_score(estimator = tpot.fitted_pipeline_, X = X, y = Y, cv = KFold(n_splits = 3, random_state = 100), n_jobs = -1, \
                        scoring = "neg_mean_squared_error")
np.mean([np.sqrt(-x) for x in score])