Scikit-learn 中的分类测试，ValueError: setting an array element with a sequence答案

【问题标题】：Classification test in Scikit-learn, ValueError: setting an array element with a sequenceScikit-learn 中的分类测试，ValueError: setting an array element with a sequence
【发布时间】：2014-05-25 08:44:48
【问题描述】：

使用tutorial on multiclass adaboost，我正在尝试对一些具有两个类的图像进行分类（但如果问题是二进制的，我认为该算法不应该工作）。然后我将扩展我的示例以包含其他类。

我目前的测试很小，总共只有 17 张图片，10 张用于训练，7 张用于测试。

现在我有两个课程：0: no vehicle, 1: vehicle present 我使用整数标签是因为根据上面链接中的示例，训练数据由基于整数的标签组成。

我只对提供的示例进行了一些编辑，以包含我自己的图像文件，但出现错误。

Traceback (most recent call last):
  File "C:\Users\app\Documents\Python Scripts\carclassify.py", line 66, in <module>
    bdt_discrete.fit(X_train, y_train)
  File "C:\Users\app\Anaconda\lib\site-packages\sklearn\ensemble\weight_boosting.py", line 389, in fit
    return super(AdaBoostClassifier, self).fit(X, y, sample_weight)
  File "C:\Users\app\Anaconda\lib\site-packages\sklearn\ensemble\weight_boosting.py", line 99, in fit
    X = np.ascontiguousarray(array2d(X), dtype=DTYPE)
  File "C:\Users\app\Anaconda\lib\site-packages\numpy\core\numeric.py", line 408, in ascontiguousarray
    return array(a, dtype, copy=False, order='C', ndmin=1)
ValueError: setting an array element with a sequence.

以下是我的代码，改编自 scikit-learn 网站上的示例：

f = open("PATH_TO_SAMPLES\\samples.txt",'r')
out = f.read().splitlines()
import numpy as np

imgs = []
tmp_hogs = []
# 13 of the images are with vehicles, 4 are without
labels = [1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0]

for file in out:
        filepath = "C:\PATH_TO_SAMPLE_IMAGES\\" + file
        curr_img = color.rgb2gray(io.imread(filepath))
        imgs.append(resize(curr_img,(60,40)))
        fd, hog_image = hog(curr_img, orientations=8, pixels_per_cell=(16, 16),
                 cells_per_block=(1, 1), visualise=True)
        tmp_hogs.append(fd) 

img_hogs = np.array(tmp_hogs)
n_split = 10
X_train, X_test = img_hogs[:n_split], X[n_split:] # all first ten images with vehicles
y_train, y_test = labels[:n_split], labels[n_split:] # 3 images with vehicles, 4 without

#now all the code below is straight off the example on scikit-learn's website

bdt_real = AdaBoostClassifier(
    DecisionTreeClassifier(max_depth=2),
    n_estimators=600,
    learning_rate=1)

bdt_discrete = AdaBoostClassifier(
    DecisionTreeClassifier(max_depth=2),
    n_estimators=600,
    learning_rate=1.5,
    algorithm="SAMME")

bdt_real.fit(X_train, y_train)
bdt_discrete.fit(X_train, y_train)

real_test_errors = []
discrete_test_errors = []

for real_test_predict, discrete_train_predict in zip(
        bdt_real.staged_predict(X_test), bdt_discrete.staged_predict(X_test)):
    real_test_errors.append(
        1. - accuracy_score(real_test_predict, y_test))
    discrete_test_errors.append(
        1. - accuracy_score(discrete_train_predict, y_test))

n_trees = xrange(1, len(bdt_discrete) + 1)

pl.figure(figsize=(15, 5))

pl.subplot(131)
pl.plot(n_trees, discrete_test_errors, c='black', label='SAMME')
pl.plot(n_trees, real_test_errors, c='black',
        linestyle='dashed', label='SAMME.R')
pl.legend()
pl.ylim(0.18, 0.62)
pl.ylabel('Test Error')
pl.xlabel('Number of Trees')

pl.subplot(132)
pl.plot(n_trees, bdt_discrete.estimator_errors_, "b", label='SAMME', alpha=.5)
pl.plot(n_trees, bdt_real.estimator_errors_, "r", label='SAMME.R', alpha=.5)
pl.legend()
pl.ylabel('Error')
pl.xlabel('Number of Trees')
pl.ylim((.2,
        max(bdt_real.estimator_errors_.max(),
            bdt_discrete.estimator_errors_.max()) * 1.2))
pl.xlim((-20, len(bdt_discrete) + 20))

pl.subplot(133)
pl.plot(n_trees, bdt_discrete.estimator_weights_, "b", label='SAMME')
pl.legend()
pl.ylabel('Weight')
pl.xlabel('Number of Trees')
pl.ylim((0, bdt_discrete.estimator_weights_.max() * 1.2))
pl.xlim((-20, len(bdt_discrete) + 20))

# prevent overlapping y-axis labels
pl.subplots_adjust(wspace=0.25)
pl.show()

编辑

我输入了

print tmp_hogs

输出是这样的：

[ array([ 0.27621208,  0.11038658,  0.10698133, ...,  0.08661556,        0.04612063,  0.0280782 ]), 
        array([  0.00000000e+00,   0.00000000e+00,   0.00000000e+00, ..., -1.29909838e-15,  -7.01780982e-17,  -1.24900943e-15]), 
        array([ 0.0503603 ,  0.1497235 ,  0.2372957 , ...,  0.07249325, 0.04545541,  0.00903818]), 
        array([ 0.27299191,  0.13122109,  0.0719268 , ...,  0.0848522 ,  0.04789403,  0.01387038]), 
        array([  0.00000000e+00,   0.00000000e+00,   0.00000000e+00, ...,  3.32140617e-17,  -6.58924128e-17,  -6.23567224e-16]), 
        array([ 0.37431874,  0.18094303,  0.01219871, ...,  0.06501856, 0.04855516,  0.02439321]), 
        array([ 0.41087302,  0.16478851,  0.03396399, ...,  0.09511273, 0.04077713,  0.03945513]), 
        array([ 0.17753915,  0.07025565,  0.09136909, ...,  0.03396507, 0.01379266,  0.01645722]), 
        array([ 0.40605587,  0.05915388,  0.03767763, ...,  0.08981079, 0.05452031,  0.01725399]), 
        array([ 0.        ,  0.        ,  0.        , ...,  0.00579303, 0.02053979,  0.0019091 ]), 
        array([ 0.31550735,  0.11988131,  0.07716529, ...,  0.09815158, 0.03058497,  0.02236517]), 
        array([  0.00000000e+00,   0.00000000e+00,   0.00000000e+00, ..., -3.51175682e-16,   1.31619418e-03,   2.86127901e-16]), 
        array([ 0.21381704,  0.22352378,  0.11568828, ...,  0.06311083, 0.02696666,  0.00402261]), 
        array([ 0.17480064,  0.1469145 ,  0.16336016, ...,  0.05614001, 0.03244093,  0.00524034]), 
        array([ 0.        ,  0.        ,  0.        , ...,  0.03089959, 0.00509584,  0.00247698]), 
        array([ 0.04711166,  0.0218663 ,  0.05316   , ...,  0.04214594, 0.04892439,  0.25840958]), 
        array([ 0.05357464,  0.00530857,  0.07162301, ...,  0.06802692, 0.08331959,  0.26619977])]

然后我跑了

print img_hogs

输出是：

[ array([ 0.27621208,  0.11038658,  0.10698133, ...,  0.08661556, 0.04612063,  0.0280782 ])
 array([  0.00000000e+00,   0.00000000e+00,   0.00000000e+00, ..., -1.29909838e-15,  -7.01780982e-17,  -1.24900943e-15])
 array([ 0.0503603 ,  0.1497235 ,  0.2372957 , ...,  0.07249325, 0.04545541,  0.00903818])
 array([ 0.27299191,  0.13122109,  0.0719268 , ...,  0.0848522 , 0.04789403,  0.01387038])
 array([  0.00000000e+00,   0.00000000e+00,   0.00000000e+00, ..., 3.32140617e-17,  -6.58924128e-17,  -6.23567224e-16])
 array([ 0.37431874,  0.18094303,  0.01219871, ...,  0.06501856, 0.04855516,  0.02439321])
 array([ 0.41087302,  0.16478851,  0.03396399, ...,  0.09511273, 0.04077713,  0.03945513])
 array([ 0.17753915,  0.07025565,  0.09136909, ...,  0.03396507, 0.01379266,  0.01645722])
 array([ 0.40605587,  0.05915388,  0.03767763, ...,  0.08981079, 0.05452031,  0.01725399])
 array([ 0.        ,  0.        ,  0.        , ...,  0.00579303, 0.02053979,  0.0019091 ])
 array([ 0.31550735,  0.11988131,  0.07716529, ...,  0.09815158, 0.03058497,  0.02236517])
 array([  0.00000000e+00,   0.00000000e+00,   0.00000000e+00, ..., -3.51175682e-16,   1.31619418e-03,   2.86127901e-16])
 array([ 0.21381704,  0.22352378,  0.11568828, ...,  0.06311083, 0.02696666,  0.00402261])
 array([ 0.17480064,  0.1469145 ,  0.16336016, ...,  0.05614001, 0.03244093,  0.00524034])
 array([ 0.        ,  0.        ,  0.        , ...,  0.03089959, 0.00509584,  0.00247698])
 array([ 0.04711166,  0.0218663 ,  0.05316   , ...,  0.04214594, 0.04892439,  0.25840958])
 array([ 0.05357464,  0.00530857,  0.07162301, ...,  0.06802692, 0.08331959,  0.26619977])]

【问题讨论】：

完全独立于错误：17 个样本显然不足以做任何有意义的事情。你为什么不下载一个标准的图像数据库并尝试一下呢？ Caltech101 是一个简单、组织良好的项目。
你能展示一下tmp_hogs 和img_hogs 的样子吗？
当然！我已经编辑了问题以包括最后的输出。
第二个输出看起来不正确（它是第一个的副本）。它应该说 array([... ... ... ...], dtype= ...)
我刚试过 img_hogs = np.array(tmp_hogs, dtype =float)，但它给出了同样的错误，而且实际上在这一行。

标签： python numpy scikit-learn

【解决方案1】：

尝试：

imgs = []
tmp_hogs = np.zeros((17, 256))
# 13 of the images are with vehicles, 4 are without
labels = [1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0]

i = 0
for file in out:
        filepath = "C:\PATH_TO_SAMPLE_IMAGES\\" + file
        curr_img = color.rgb2gray(io.imread(filepath))
        imgs.append(resize(curr_img,(60,40)))
        fd, hog_image = hog(curr_img, orientations=8, pixels_per_cell=(16, 16),
                 cells_per_block=(1, 1), visualise=True)
        tmp_hogs[i,:] = fd
        i+=1

img_hogs = tmp_hogs

【讨论】：

刚试过这个，但在 tmp_hogs[i,:]=fd 行，我得到以下错误：ValueError: could not broadcast input array from shape (1728) 变成形状 (256)。所以我调整了 tmp_hogs 声明并给了它 1728 列，然后我又得到了同样的错误，这一次说它不能“从 1728 年到 2080 年广播”。所以我猜这意味着我的 tmp_hogs 在每一行中没有相同数量的列？但是我将所有图像的大小调整为 60,40！那怎么可能呢？
限制 hog 特征的数量。它不一致
怎么样？我认为将图像大小调整为相同大小会限制这一点。