如何在scikit-learn中创建我自己的数据集？

Question

我想创建自己的数据集，并在scikit-learn中使用它。 Scikit-learn有一些数据集，如“波士顿住房数据集”和“波士顿住房数据集”。 （.csv），用户可以通过以下方式使用它：

from sklearn import datasets 
boston = datasets.load_boston()

以下代码可以获取此数据集的data和target：

X = boston.data
y = boston.target

问题是如何创建我自己的数据集并可以这样使用？ 任何答案都表示赞赏，谢谢！

Answer 1

这是实现目标的快捷方式：

my_datasets.py

import numpy as np
import csv
from sklearn.datasets.base import Bunch

def load_my_fancy_dataset():
    with open('my_fancy_dataset.csv') as csv_file:
        data_file = csv.reader(csv_file)
        temp = next(data_file)
        n_samples = int(temp[0])
        n_features = int(temp[1])
        data = np.empty((n_samples, n_features))
        target = np.empty((n_samples,), dtype=np.int)

        for i, sample in enumerate(data_file):
            data[i] = np.asarray(sample[:-1], dtype=np.float64)
            target[i] = np.asarray(sample[-1], dtype=np.int)

    return Bunch(data=data, target=target)

my_fancy_dataset.csv

5,3,first_feat,second_feat,third_feat
5.9,1203,0.69,2
7.2,902,0.52,0
6.3,143,0.44,1
-2.6,291,0.15,1
1.8,486,0.37,0

演示

In [12]: import my_datasets

In [13]: mfd = my_datasets.load_my_fancy_dataset()

In [14]: X = mfd.data

In [15]: y = mfd.target

In [16]: X
Out[16]: 
array([[  5.90000000e+00,   1.20300000e+03,   6.90000000e-01],
       [  7.20000000e+00,   9.02000000e+02,   5.20000000e-01],
       [  6.30000000e+00,   1.43000000e+02,   4.40000000e-01],
       [ -2.60000000e+00,   2.91000000e+02,   1.50000000e-01],
       [  1.80000000e+00,   4.86000000e+02,   3.70000000e-01]])

In [17]: y
Out[17]: array([2, 0, 1, 1, 0])

Answer 2

假设您已将数据作为2D数组加载到内存中，可以使用OneHotEncoder来实现这一目的的简便方法：

from sklearn import svm
from sklearn.preprocessing import OneHotEncoder

data = [['Male', 1, True], ['Female', 3, True], ['Female', 2, False], ['Male', 2, False]]
y = [0, 1, 1, 1]  # expected outputs
enc = OneHotEncoder(drop='if_binary')  # create encoder obj that drops unneeded columns on binary inputs
X = enc.fit_transform(data).toarray()  # vectorize input data
clf = svm.SVC(gamma=0.001, C=100.)  # create classifier obj
clf.fit(X[:-1], y[:-1])  # fit model using training data (all but last entry)

out = clf.predict(X[-1:])  # resulting prediction (last entry only)