管道未将standardscaler正确用于自定义模型

Question

我正试图用一个标准的sklearn scaler和一个自写模型类建立一个管道。在写costum课程时，我遵循了sklearn documentation。管道应该是这样的

model_piped = make_pipeline(StandardScaler(), model)

问题在于，在拟合管道肋模型后，它没有使用StandardScaler的逆_变换方法来重新缩放其预测。因此，Costam模型只对输入进行缩放。当然，管道适用于标准的sklearn模型，如Lasso回归

一个有效的例子：

import numpy as np

from sklearn.linear_model import Lasso
from sklearn.pipeline import make_pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.utils.validation import check_X_y, check_array, check_is_fitted

from matplotlib import pyplot as plt

# =============================================================================
# An example costum model
# =============================================================================
class mean_stacker():
    def __init__(self, desc = 'Simple Mean Stacker'):
        self.description = desc

    def predict(self, X):
        check_is_fitted(self, ['coef_'])
        X = check_array(X)
        return np.mean(X, axis = 1)

    def fit(self, X, y):
        X, y = check_X_y(X, y)
        self.coef_ = 'mean'
        return self

# =============================================================================
# test data  
# =============================================================================
yhats = np.array([[11.64543231, 11.49851957, 11.89059499, 11.77613068],
       [12.0166365 , 12.18640595, 11.89059499, 12.03356647],
       [11.91435714, 12.00392321, 11.89059499, 12.00279713],
       [11.74216858, 11.57740889, 11.89059499, 11.57306004],
       [11.9827991 , 12.09409814, 11.89059499, 12.14146709],
       [11.64009661, 11.55337117, 11.89059499, 11.539958  ],
       [11.8658174 , 11.93479133, 11.89059499, 11.88695717],
       [11.53478821, 11.24788878, 11.89059499, 11.47217846],
       [12.03600978, 12.16789499, 11.89059499, 12.09874916],
       [12.07294432, 12.20473012, 11.89059499, 12.20545864],
       [11.62189652, 11.34157305, 11.89059499, 11.4359684 ],
       [11.5167136 , 11.66579694, 11.89059499, 11.58799839]])

ytrue = np.array(
        [[11.6524265 ],
         [11.89470562],
         [12.12673719],
         [12.7966    ],
         [11.86452555],
         [11.85743673],
         [11.45650325],
         [11.96433224],
         [12.33647352],
         [11.96876678],
         [11.24377724],
         [11.2209676 ]])

# =============================================================================
# Define set of stacker models, wrap them into pipelines and fit
# =============================================================================
stacker_models =[ 
                 Lasso(alpha = 0.0005, random_state = 4),
                 mean_stacker()
                 ]
stacker = []
for model in stacker_models:
    stkr = make_pipeline(StandardScaler(), model)
    stkr.fit(yhats, ytrue)
    stacker.append(stkr)  

# =============================================================================
# plot predicts from models
# the costum model does not rescale the predicts correctly 
# =============================================================================
plt.plot(ytrue, label = 'ytrue')
plt.plot(stacker[0].predict(yhats), label = 'yLasso')
plt.plot(stacker[1].predict(yhats), label = 'yMean')
plt.legend()

有没有关于如何修复Costam模型的提示