我将 sklearn 的 FeatureHasher 用于 DecisionTreeRegressor。现在我如何解码回归器的预测结果？答案

【问题标题】：I use FeatureHasher from sklearn for DecisionTreeRegressor. Now how do i decode the predicted results from regressor?我将 sklearn 的 FeatureHasher 用于 DecisionTreeRegressor。现在我如何解码回归器的预测结果？
【发布时间】：2018-07-14 10:59:05
【问题描述】：

dataset = pd.read_csv('ll.csv',index_col=0)
dataset = dataset.dropna(axis=0)

# features or independent variables
x = pd.DataFrame()
x['Skills'] = dataset['Skills']
x['Location'] = dataset['Location']
x['Industry'] = dataset['Industry']
x['Experience'] = dataset['Experience']

# applying hashing
x_hash = copy.copy(x)

for i in range(x_hash.shape[1]):
    x_hash.iloc[:,i] = x_hash.iloc[:,i].astype('str')

x_hash = h.transform(x_hash.values)

#Dependent Variable
y=pd.DataFrame()

y['Functional Area'] = dataset['Functional Area']
y_hash = copy.copy(y)

for i in range(y_hash.shape[1]):
    y_hash.iloc[:,i] = y_hash.iloc[:,i].astype('str')

y_hash = h.transform(y_hash.values)

# Regressor
regressor = DecisionTreeRegressor(random_state=0)

ll = regressor.fit(x_hash.toarray(),y_hash.toarray())

# For predicting input features
input_df = pd.DataFrame()

input_df['Skills'] = ['Illustrator']

input_df['Experience'] = ['1-6']

input_df['Industry'] = ['IT - Software Services']

input_df['Location'] = ['Cairo-Egypt']

input_df_hash = copy.copy(input_df)

for i in range(input_df_hash.shape[1]):
    input_df_hash.iloc[:,i] = input_df_hash.iloc[:,i].astype('str')


input_df_hash = h.transform(input_df_hash.values)

sss=regressor.predict(input_df_hash.toarray())

【问题讨论】：

标签： python scikit-learn random-forest feature-extraction

【解决方案1】：

我认为你需要稍微修改一下。

一个可能的建议是手动散列文档中的所有字符串并查看它们对应的列（特征）。

一个最小的例子：

from sklearn.feature_extraction import FeatureHasher, _hashing
import numpy as np

def hasher(string, n_features):
    """Hash a single string"""

    res = _hashing.transform([[(string, 1)]], dtype=int, n_features=n_features)
    return res[0][0]


n_features = 10

h = FeatureHasher(n_features=n_features)
D = [{'dog': 1, 'cat': 2, 'elephant': 4}, {'dog': 2, 'run': 5}, {'human': 4, 'desk': 10}]

word_to_ix = {word: hasher(word, n_features) for word in set().union(*(d.keys() for d in D))}
columns = ['-'.join([i for i in word_to_ix.keys() if word_to_ix[i] == ix]) for ix in range(n_features)]

f = h.transform(D)

print('Transformed features:\n', f.toarray())
print('Word to ix dictionary:\n', word_to_ix)
print('Columns:\n', columns)

输出：

Transformed features:
 [[ 0.  0. -4. -1.  0.  0.  0.  0.  0.  2.]
 [ 0.  0.  0. -2. -5.  0.  0.  0.  0.  0.]
 [ 0.  0.  0.  0.  0.  0.  4.  0.  0. 10.]]
Word to ix dictionary:
 {'dog': 3, 'desk': 9, 'elephant': 2, 'human': 6, 'run': 4, 'cat': 9}
Columns:
 ['', '', 'elephant', 'dog', 'run', '', 'human', '', '', 'desk-cat']

请注意，这种方法可以处理可能的冲突。

【讨论】：