原文：https://blog.csdn.net/angel_hben/article/details/84341421

import numpy as np

import tensorflow as tf

p.random.seed(0)

x = np.random.sample((11,2)) # make a dataset from a numpy array

print(x)

dataset = tf.data.Dataset.from_tensor_slices(x) ####

#####最好用 dataset = tf.data.Dataset.from_generator(self.generator, (tf.float32, tf.int32,tf.int32, tf.string))，使用例子如下：

‘’‘

def generator(self):

                    for index in range(len(self.data_list)):

                            file_basename_image,file_basename_label = self.data_list[index]

                            image_path = os.path.join(self.data_dir, file_basename_image)

                            label_path= os.path.join(self.data_dir, file_basename_label)

                            image= self.read_data(image_path)

                            label = self.read_data(label_path)

                            label_pixel,label=self.label_preprocess(label)

                            image = (np.array(image[:, :, np.newaxis]))

                            label_pixel = (np.array(label_pixel[:, :, np.newaxis]))

                            yield image, label_pixel,label, file_basename_image

’‘’

dataset = dataset.shuffle(2)#将数据打乱，数值越大，混乱程度越大

dataset = dataset.batch(4)#按照顺序取出4行数据，最后一次输出可能小于batch

dataset = dataset.repeat()#数据集重复了指定次数

# repeat()在batch操作输出完毕后再执行,若在之前，相当于先把整个数据集复制两次 #为了配合输出次数，一般默认repeat()空

# create the iterator

iter = dataset.make_one_shot_iterator()

el = iter.get_next()

with tf.Session() as sess:

         for i in range(6):

                  value = sess.run(el)

                  print(value)

---------------------