Question

我正在尝试创建一个稀疏类别的keras模型，但出现以下错误：ValueError：设置具有序列的数组元素。

我在这里浏览了关于同一错误的许多主题，并尝试了提供的一些解决方案，但是都没有运气。

一种解决方案是检查以确保输入数组的大小相同（solution here）。我检查了以下代码：

for i in range(len(pre_x)):
    shape_x = pre_x[0].shape
        if pre_x[i].shape != shape_x:
            print(pre_x[i].shape)
            print(i)

pre-x是输入x。不，我没有打印出来，所以这不是错误。

我也曾在声明数组时尝试指定数据类型，例如，当我将x数据中的某些列表转换为数组时，我使用了以下代码：

pre_x = np.array(pre_x, dtype=object)

同样，错误仍然存在，因此这不是错误。另外，我尝试对y变量进行一次热编码，使之仅是分类熵，而不是稀疏的分类熵，但是还是有同样的错误。

这是我的简化代码：

import matplotlib.pyplot as plt
import numpy as np
import scipy.io.wavfile
import librosa
import librosa.display
from scipy import misc

from sklearn.model_selection import train_test_split
from sklearn.metrics import confusion_matrix
from keras.utils.np_utils import to_categorical
from keras.models import Sequential
from keras.layers import Dense, Dropout, Conv2D, MaxPool2D
from keras.preprocessing.image import ImageDataGenerator

def image_to_numpy_array(img_location):
    img = (misc.imread(img_location, flatten=True))/255
    img = img.tolist()
    for i in range(49,126):
        for n in range(len(img[0])):
            img[i][n] = [img[i][n]]
    img = np.array(img, dtype=object)
    return img[50:125]

def genres_to_sectionlist(genres):
    for i in range(len(genres)):
        fail_count = 0
        song_count = 0
        print("Currently uploading genre: " + genres[i])
        while fail_count < 2:
            section_count = 0
            Temp_sections = []
            #fail_count = 0
            while True:
                try:
                    section = [i, image_to_numpy_array('TempImage\\' + genres[i] + str(song_count) + '_' + str(section_count) + '.jpg')]
                    Temp_sections.append(section)
                    fail_count = 0
                except:
                    fail_count += 1
                    break
                section_count += 1
            if len(Temp_sections) > 1:
                All_sections.append(Temp_sections)
            print("Song number: %i" % song_count)
            song_count += 1
    return All_sections

All_sections = []
Temp_sections = []
genres = ["Classic_Doom", "2W_Black_Metal", "Cavern_Death"]
# genres = ["Cavern_Death", "2W_Black_Metal"]

All_sections = genres_to_sectionlist(genres)

for i in range(len(All_sections)):
    print(len(All_sections[i]))

for i in range(len(All_sections)):
    for j in range(len(All_sections[i])):
        All_sections[i][j][1] = np.expand_dims(All_sections[i][j][1], axis=2)

for i in range(len(All_sections)):

    full_data = []
    Test_sections = All_sections[i]
    for n in range(len(All_sections)):
        if i != n:
            full_data += All_sections[n]

    print("Training data length: " + str(len(full_data)))
    print("Test data length: " + str(len(Test_sections)))

    pre_x = []
    pre_y = []
    test_x = []
    test_y = []


    for j in range(len(full_data)):
        pre_x.append(full_data[j][1])
        pre_y.append(full_data[j][0])

    for j in range(len(Test_sections)):
        test_x.append(Test_sections[j][1])
        test_y.append(Test_sections[j][0])

    pre_x = np.array(pre_x, dtype=object)
    test_x = np.array(test_x, dtype=object)

    model = Sequential()

    model.add(Conv2D(32, (3, 3), activation='relu', input_shape=(75, 192, 1)))
    model.add(MaxPool2D((2, 2)))
    model.add(Conv2D(64, (3, 3), activation='relu'))
    model.add(MaxPool2D((2, 2)))
    model.add(Conv2D(128, (3, 3), activation='relu'))
    model.add(MaxPool2D((2, 2)))
    model.add(Conv2D(128, (3, 3), activation='relu'))
    model.add(MaxPool2D((2, 2)))
    model.add(Flatten())
    model.add(Dropout(0.5))
    model.add(Dense(512, activation='relu'))
    model.add(Dense(1, activation='softmax'))

    print("TESTING...")
    for i in range(len(pre_y)):
        shape_x = pre_y[0].shape
        if pre_y[i].shape != shape_x:
            print(pre_y[i].shape)
            print(i)

    datagen = ImageDataGenerator()
    model.compile(loss='sparse_categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
    hist = model.fit_generator(datagen.flow(pre_x, pre_y, batch_size=8), steps_per_epoch=500, epochs=10, verbose=2, validation_data=(test_x, test_y))

有人怀疑为什么会发生此错误吗？有什么我可以尝试的吗？

编辑：错误输出如下：

Traceback (most recent call last):
  File "Analyzer_Keras.py", line 158, in <module>
    hist = model.fit_generator(datagen.flow(pre_x, pre_y, batch_size=8), steps_per_epoch=500, epochs=10, verbose=2, validation_data=(test_x, test_y))
  File "C:\Users\aibow\AppData\Local\Programs\Python\Python37\lib\site-packages\keras_preprocessing\image\image_data_generator.py", line 430, in flow
    subset=subset
  File "C:\Users\aibow\AppData\Local\Programs\Python\Python37\lib\site-packages\keras_preprocessing\image\numpy_array_iterator.py", line 112, in __init__
    self.x = np.asarray(x, dtype=self.dtype)
  File "C:\Users\aibow\AppData\Local\Programs\Python\Python37\lib\site-packages\numpy\core\numeric.py", line 538, in asarray
    return array(a, dtype, copy=False, order=order)
ValueError: setting an array element with a sequence.

编辑2：我已经检查了Github上的numpy asarray源代码（发生错误的地方），我唯一能真正推断出的就是它正在尝试从x数据创建数组，这已经是数组，因此它不应引发错误。有什么我要忽略的吗？

在Keras模型中使用X输入获取ValueError

0 个答案: