python - scikit拟合数据错误

Question

我正在尝试实现一种学习算法来预测图像的目标值是 1 还是 0。首先，我的目标值是这样设置的……

real = [1] * len(images)
fake = [0] * len(fake_images)

total_target = real + fake
total_target = numpy.array(total_target)

>>> [1 1 1 ... 0 0 0 0]

接下来，我将图像列表转换为 numpy 数组的 numpy 数组。所以我将每个图像存储为 numpy 数组...

training_set = []
for image in total_images:
    im = image.convert("L")
    dataset = numpy.asarray(im)
    training_set.append(dataset)
training_set = numpy.array(training_set)

因此 training_set 包含图像。 training_set的顺序对应于total_target的顺序，所以training_set中的第一个图像对应于total_target中的第一个值 在上面的示例中为 1。

接下来我展平训练集...

n_samples = len(training_set)
data = training_set.reshape((n_samples, -1))

现在我把它传递给下面...

classifier = svm.SVC(gamma=0.001)
classifier.fit(data[:n_samples-1], total_target[:n_samples-1])

我没有包括最后一张图片及其各自的值，因为这是我想要预测的值...

expected = total_target[-1]
predicted = classifier.predict(data[-1])

当我运行所有这些时，出现以下错误...

DeprecationWarning: Passing 1d arrays as data is deprecated in 0.17 and willraise ValueError in 0.19. Reshape your data either using X.reshape(-1, 1) if your data has a single feature or X.reshape(1, -1) if it contains a single sample. DeprecationWarning)

好的，根据错误，我的 total_target 格式错误，所以我添加以下内容...

total_target = numpy.array(total_target).reshape(-1, 1)

我运行它，现在我得到以下错误

DataConversionWarning: A column-vector y was passed when a 1d array was expected. Please change the shape of y to (n_samples, ), for example using ravel(). y_ = column_or_1d(y, warn=True)

C:\Users\Eric\Anaconda2\lib\site-packages\sklearn\utils\validation.py:386: DeprecationWarning: Passing 1d arrays as data is deprecated in 0.17 and willraise ValueError in 0.19. Reshape your data either using X.reshape(-1, 1) if your data has a single feature or X.reshape(1, -1) if it contains a single sample. DeprecationWarning)

我尝试在 total_target 上使用 ravel() 但它只是让我回到之前的错误。我认为我的格式有误我对 numpy 数组还很陌生。

Answer 1

Numpy 的 atleast_2d让代码工作。

让我们首先生成一些模拟数据，即 800 行 x 1200 列的 5 个真实和 5 个伪造的 8 位图像:

In [111]: import numpy as np

In [112]: real, fake = 5, 5

In [113]: rows, cols = 800, 1200

In [114]: bits = 8

In [115]: target = np.hstack([np.ones(real), np.zeros(fake)])

In [116]: np.random.seed(2017)

In [117]: images = np.random.randint(2**bits, size=(real + fake, rows, cols))

In [118]: data = images.reshape(images.shape[0], -1)

In [119]: data
Out[119]: 
array([[ 59,   9, 198, ..., 189, 201,  38],
       [150, 251, 145, ...,  95, 214, 175],
       [156, 212, 220, ..., 179,  63,  48],
       ..., 
       [ 25,  94, 108, ..., 159, 144, 216],
       [179, 103, 217, ...,  92, 219,  34],
       [198, 209, 177, ...,   6,   4, 144]])

In [120]: data.shape
Out[120]: (10L, 960000L)

然后我们使用除最后一张图像之外的所有图像来训练分类器:

In [121]: from sklearn import svm

In [122]: classifier = svm.SVC(gamma=0.001)

In [123]: classifier.fit(data[:-1], target[:-1])
Out[123]: 
SVC(C=1.0, cache_size=200, class_weight=None, coef0=0.0,
  decision_function_shape=None, degree=3, gamma=0.001, kernel='rbf',
  max_iter=-1, probability=False, random_state=None, shrinking=True,
  tol=0.001, verbose=False)

如果您现在尝试通过 classifier.predict(data[-1]) 对最后一张图像进行分类，sklearn 会报错。为了让 sklearn 开心，您只需要确保测试数据是二维的，如下所示:

In [124]: classifier.predict(np.atleast_2d(data[-1]))
Out[124]: array([ 1.])