http://scikit-learn.org/stable/modules/feature_extraction.html

帶病在網吧裏。

。。。。。

寫。求支持。

。。

1、首先澄清兩個概念：特征提取和特征選擇（

Feature extraction is very different from Feature selection

the former consists in transforming arbitrary data, such as text or images, into numerical features usable for machine learning. The latter is a machine learning technique applied on these features（從已經提取的特征中選擇更好的特征）.

以下分為四大部分來講。主要還是4、text feature extraction

2、loading features form dicts

class DictVectorizer。舉個樣例就好：

>>> measurements = [
...     {‘city‘: ‘Dubai‘, ‘temperature‘: 33.},
...     {‘city‘: ‘London‘, ‘temperature‘: 12.},
...     {‘city‘: ‘San Fransisco‘, ‘temperature‘: 18.},
... ]
 

>>> from sklearn.feature_extraction import DictVectorizer
>>> vec = DictVectorizer()
>>> vec.fit_transform(measurements).toarray()
array([[  1.,   0.,   0.,  33.],
       [  0.,   1.,   0.,  12.],
       [  0.,   0.,   1.,  18.]])
>>> vec.get_feature_names()
[‘city=Dubai‘, ‘city=London‘, ‘city=San Fransisco‘, ‘temperature‘]
 

>>> pos_window = [
...     {
...         ‘word-2‘: ‘the‘,
...         ‘pos-2‘: ‘DT‘,
...         ‘word-1‘: ‘cat‘,
...         ‘pos-1‘: ‘NN‘,
...         ‘word+1‘: ‘on‘,
...         ‘pos+1‘: ‘PP‘,
...     },
...     # in a real application one would extract many such dictionaries
... ]

上面的PoS特征就能夠vectorized into a sparse two-dimensional matrix suitable for feeding into a classifier (maybe after being piped into a text.TfidfTransformer for normalization):

>>> vec = DictVectorizer()
>>> pos_vectorized = vec.fit_transform(pos_window)
>>> pos_vectorized                
<1x6 sparse matrix of type ‘<... ‘numpy.float64‘>‘
    with 6 stored elements in Compressed Sparse ... format>
>>> pos_vectorized.toarray()
array([[ 1.,  1.,  1.,  1.,  1.,  1.]])
>>> vec.get_feature_names()
[‘pos+1=PP‘, ‘pos-1=NN‘, ‘pos-2=DT‘, ‘word+1=on‘, ‘word-1=cat‘, ‘word-2=the‘]

3、feature hashing

The class FeatureHasher is a high-speed, low-memory vectorizer that uses a technique known as feature hashing, or the “hashing trick”.

因為hash。所以僅僅保存feature的interger index。而不保存原來feature的string名字。所以沒有inverse_transform方法。

FeatureHasher 接收dict對，即 (feature, value) 對，或者strings，由構造函數的參數input_type決定.結果是scipy.sparse matrix。假設是strings，則value默認取1，比如 [‘feat1‘, ‘feat2‘, ‘feat2‘] 被解釋為[(‘feat1‘, 1), (‘feat2‘, 2)].

4、text feature extraction

由於內容太多，分開寫了。參考著篇博客：http://blog.csdn.net/mmc2015/article/details/46997379

5、image feature extraction

提取部分圖片（Patch extraction）：

The extract_patches_2d function從圖片中提取小塊，存儲成two-dimensional array, or three-dimensional with color information along the third axis. 使用reconstruct_from_patches_2d. 可以將全部的小塊重構成原圖：

>>> import numpy as np
>>> from sklearn.feature_extraction import image

>>> one_image = np.arange(4 * 4 * 3).reshape((4, 4, 3))
>>> one_image[:, :, 0]  # R channel of a fake RGB picture
array([[ 0,  3,  6,  9],
       [12, 15, 18, 21],
       [24, 27, 30, 33],
       [36, 39, 42, 45]])

>>> patches = image.extract_patches_2d(one_image, (2, 2), max_patches=2,
...     random_state=0)
>>> patches.shape
(2, 2, 2, 3)
>>> patches[:, :, :, 0]
array([[[ 0,  3],
        [12, 15]],

       [[15, 18],
        [27, 30]]])
>>> patches = image.extract_patches_2d(one_image, (2, 2))
>>> patches.shape
(9, 2, 2, 3)
>>> patches[4, :, :, 0]
array([[15, 18],
       [27, 30]])

>>> reconstructed = image.reconstruct_from_patches_2d(patches, (4, 4, 3))
>>> np.testing.assert_array_equal(one_image, reconstructed)

The PatchExtractor class和 extract_patches_2d,一樣，僅僅只是能夠同一時候接受多個圖片作為輸入：

>>> five_images = np.arange(5 * 4 * 4 * 3).reshape(5, 4, 4, 3)
>>> patches = image.PatchExtractor((2, 2)).transform(five_images)
>>> patches.shape
(45, 2, 2, 3)

圖片像素的連接（Connectivity graph of an image）：

主要是依據像素的區別來推斷圖片的每兩個像素點是否連接。

。。

。

。

The function img_to_graph returns such a matrix from a 2D or 3D image. Similarly, grid_to_graph build a connectivity matrix for images given the shape of these image.

這有個直觀的樣例：http://scikit-learn.org/stable/auto_examples/cluster/plot_lena_ward_segmentation.html#example-cluster-plot-lena-ward-segmentation-py

頭疼。。。。

碎覺。

。。

scikit-learn：4.2. Feature extraction（特征提取，不是特征選擇）