构建深度 CNN 时出现 InvalidArgumentError [重复]

Question

我是 TensorFlow 和 python 的新手... 我正在尝试为细胞图像分类 Hep-2 数据集构建一个深度 CNN。数据集由13596 图像组成，我使用8701 图像作为CNN 的训练数据。另外，我有 .CSV 文件，其中包含图像 ID 及其单元格类型。我提取了内容并使用 .CSV 文件中的 image_ID 作为我的标签。训练数据和图像 ID 均已转换为 .astype(‘float32’)。但是，不知怎的，我得到了 InvalidArgumentError，我不知道里面发生了什么。 我已经发布了我的代码和错误，任何提示或帮助将不胜感激。提前谢谢你:)

我也是 Stack Overflow 的新手。抱歉我的格式混乱。

我的代码：

from PIL import Image
import glob
import os
import numpy as np
import pandas as pd
import tensorflow as tf
from tensorflow import keras
from keras.models import Sequential
from keras.layers import Dense, Conv2D, Dropout, Flatten, MaxPooling2D
from keras.optimizers import SGD

def extract_labels(image_names, Original_Labels):
  temp = np.array([image.split('.')[0] for image in image_names])
  temp2 = np.array([j[0] for i in temp for j in Original_Labels if(int(i) == int(j[0]))])
  return temp2

def get_Labels():
  df=pd.read_csv('gt_training.csv', sep=',')
  labels = np.asarray(df)

  path = 'path..../training/'
  image_names_train = [f for f in os.listdir(path) if os.path.splitext(f)[-1] == '.png']

  return labels, image_names_train



Train_images = glob.glob('path.../training/*.png')
train_data = np.array([np.array(Image.open(fname)) for fname in Train_images])

train_data = train_data.astype('float32')
train_data /= 255

#getting labels from .csv file for training data
labels, image_names_train = get_Labels()
train_labels = extract_labels(image_names_train, labels)
train_labels = train_labels.astype('float32')
print(train_labels.shape)

train_data = train_data.reshape(train_data.shape[0],78,78,1) #reshaping into 4-Dim
input_shape = (78, 78, 1) #1 because the provided dataset is in grey scale 


#Adding pooling, dense layers to an an non-optimized empty CNN
model = Sequential()
model.add(Conv2D(6, kernel_size=(7,7),activation = tf.nn.tanh, input_shape = input_shape))
model.add(MaxPooling2D(pool_size = (2, 2)))
model.add(Conv2D(16, kernel_size=(4,4),activation = tf.nn.tanh))
model.add(MaxPooling2D(pool_size = (3, 3)))
model.add(Conv2D(32, kernel_size=(3,3),activation = tf.nn.tanh))
model.add(MaxPooling2D(pool_size = (3, 3)))
model.add(Flatten())
model.add(Dense(150, activation = tf.nn.tanh, kernel_regularizer = keras.regularizers.l2(0.00005)))
model.add(Dropout(0.5))
model.add(Dense(6, activation = tf.nn.softmax))

#setting an optimizer with a given loss function
opt = SGD(lr = 0.01, momentum = 0.9)
model.compile(optimizer = opt, loss = 'sparse_categorical_crossentropy', metrics = ['accuracy'])
model.fit(x = train_data, y = train_labels, epochs = 10, batch_size = 77)

我得到的错误信息：

six.raise_from(core._status_to_exception(e.code, message), None)

  File "<string>", line 3, in raise_from

InvalidArgumentError:  Received a label value of 13269 which is outside the valid range of [0, 6).  Label values: 8823 3208 9410 5223 8817 3799 6588 1779 1371 5017 9788 9886 3345 1815 5943 37 675 2396 4485 9528 11082 12457 13269 5488 3250 12896 13251 1854 10942 6287 6232 2944
     [[node loss_24/dense_55_loss/sparse_categorical_crossentropy/SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits (defined at C:\Users\vardh\Anaconda3\envs\tf\lib\site-packages\keras\backend\tensorflow_backend.py:3009) ]] [Op:__inference_keras_scratch_graph_676176]

Function call stack:
keras_scratch_graph

Answer 1

不知何故，我意识到我的问题与这个问题 InvalidArgumentError: Received a label value of 8825.... 有关。

来自该帖子的解决方案： @shaili 已发布，

在最后一层，例如，您使用了 model.add(Dense(1, activation='softmax'))。这里 1 将其值从 [0, 1) 限制为将其形状更改为最大输出标签。例如，您的输出来自标签 [0,7)，然后使用 model.add(Dense(7, activation='softmax'))

input_text = Input(shape=(max_len,), dtype=tf.string)
embedding = Lambda(ElmoEmbedding, output_shape=(max_len, 1024))(input_text)
x = Bidirectional(LSTM(units=512, return_sequences=True,
                   recurrent_dropout=0.2, dropout=0.2))(embedding)
x_rnn = Bidirectional(LSTM(units=512, return_sequences=True,
                       recurrent_dropout=0.2, dropout=0.2))(x)
x = add([x, x_rnn])  # residual connection to the first biLSTM
out = TimeDistributed(Dense(n_tags, activation="softmax"))(x)
Here in TimeDistributed layer n_tags is the length of tags from which I want to classify.

如果我预测一些其他数量，例如长度与 n_tags 不同的 q_tag，即假设 10 并且 n_tags 的长度为 7，并且我收到 8 作为输出标签，它将给出无效参数错误 Received a label value of 8 which is outside [0, 7) 的有效范围。

根据我的经验， 通常，该错误是由于没有要分类的分类不准确而产生的。在我的代码中，model.add(Dense(6, activation = tf.nn.softmax) 我给了6 要生成的分类类型，而不是13596。但是，这不是一个完全可以工作的代码，至少它可以让我的代码运行。