如何在 Pipeline 中使用 LabelEncoder 和随机森林答案

【问题标题】：how to use LabelEncoder and Random Forest in Pipeline如何在 Pipeline 中使用 LabelEncoder 和随机森林
【发布时间】：2019-12-06 19:23:56
【问题描述】：

我正在尝试使用来自scikit-learn 的Pipeline。目前我正在做以下事情：

在某些功能上应用LabelEncoder
构建RandomForest 回归器

代码是：

x['zipcode'] = labelencoder.fit_transform(x['zipcode'])

rfr = RandomForestRegressor(n_estimators=20, random_state=0)

rfr.fit(x, y)

我如何构建一个Pipeline 以便未来看不见的数据经历相同的转换？

【问题讨论】：

标签： python machine-learning scikit-learn random-forest

【解决方案1】：

您无需将 LabelEncoder 转换放入 sklearn Pipeline 指令中。因此，可能的解决方案是调用 LabelEncoder ，例如如下：

import numpy as np 
from sklearn.preprocessing import LabelEncoder

lbl = LabelEncoder()
lbl.fit(X)
np.save('lbl_encoder.npy', encoder.classes_)

并在需要时加载它


lbl = LabelEncoder
lbl.classes_ = np.load('lbl_encoder.npy')

【讨论】：

谢谢。但是，我计划部署最终模型，并且使用 Pipeline，您可以将模型保存为 pkl 文件，其中也将包含所有转换器。

【解决方案2】：

为了解决这个问题，我会创建自己的PipeLine。请考虑这个简单的示例，您可以根据自己的要求对其进行自定义和添加：

import copy
import pickle
from sklearn.ensemble import RandomForestRegressor
from sklearn.preprocessing import LabelEncoder
import pandas as pd
import numpy as np

class MyPipeline:
    def __init__(self, column='zipcode', n_estimators=10, random_state=0):
        # add any parameters as required
        # customize as required
        assert isinstance(column, str)
        self.column = column
        self.label_enc = LabelEncoder()
        self.model = RandomForestRegressor(n_estimators=n_estimators,
                                           random_state=random_state,
                                           n_jobs=1)

    def preprocess(self, x):
        processed_x = copy.deepcopy(x)
        processed_x[self.column] = self.label_enc.fit_transform(processed_x[self.column])
        return np.array(processed_x)

    def fit(self, x, y):
        # also customize as required
        transformed_x = self.preprocess(x)
        print("X After Transform: \n{}\n".format(transformed_x))
        self.model.fit(transformed_x, y)

    def predict(self, unseen_x):
        # also customize as required
        processed_x = copy.deepcopy(unseen_x)
        processed_x[self.column] = self.label_enc.transform(processed_x[self.column])
        print("Unseen Data After Transform: \n{}\n".format(np.array(processed_x)))
        return self.model.predict(np.array(processed_x))

测试

x = pd.DataFrame(columns=['blabla','zipcode'],
                 data=[[1, 'zipecode1'], [2,'zipecode2'], [3,'zipecode3'],
                       [4, 'zipecode4'], [5, 'zipecode5'], [6, 'zipecode6'],
                       [7, 'zipecode7'], [8, 'zipecode8'], [9, 'zipecode9']])
y = [10,20,30,40,50,60,70,80,90]

mypipeline = MyPipeline()
mypipeline.fit(x,y)

# save it for future work
with open('mypipeline.dat', 'wb') as pickle_file:
    pickle.dump(mypipeline, pickle_file)

# retrieve it
with open('mypipeline.dat', 'rb') as pickle_file:
    mypipeline_ = pickle.load(pickle_file)

# Here I am passing same x just to make sure it's doing proper transformation
result = mypipeline_.predict(x)

# the result
print("Results: {}".format(result))

输出

X After Transform: 
[[1 0]
 [2 1]
 [3 2]
 [4 3]
 [5 4]
 [6 5]
 [7 6]
 [8 7]
 [9 8]]

Unseen Data After Transform: 
[[1 0]
 [2 1]
 [3 2]
 [4 3]
 [5 4]
 [6 5]
 [7 6]
 [8 7]
 [9 8]]

Results: [12. 18. 28. 37. 47. 56. 69. 79. 82.]

【讨论】：