在学习机器学习时，通常会遇到数据集的问题，墙就是一座翻不完的大山，感谢谷歌提供的数据集的包，再也不用担心数据集的问题了。其安装也非常简单，直接pip就行

pip install tensorflow-datasets

以下罗列了tensorflow-datasets现有的数据集。

• audio

  • "groove"
  • "nsynth"

• image

  • "abstract_reasoning"
  • "caltech101"
  • "cats_vs_dogs"
  • "celeb_a"
  • "celeb_a_hq"
  • "cifar10"
  • "cifar100"
  • "cifar10_corrupted"
  • "coco2014"
  • "colorectal_histology"
  • "colorectal_histology_large"
  • "cycle_gan"
  • "diabetic_retinopathy_detection"
  • "dsprites"
  • "dtd"
  • "emnist"
  • "fashion_mnist"
  • "horses_or_humans"
  • "image_label_folder"
  • "imagenet2012"
  • "imagenet2012_corrupted"
  • "kmnist"
  • "lsun"
  • "mnist"
  • "omniglot"
  • "open_images_v4"
  • "oxford_flowers102"
  • "oxford_iiit_pet"
  • "quickdraw_bitmap"
  • "rock_paper_scissors"
  • "shapes3d"
  • "smallnorb"
  • "sun397"
  • "svhn_cropped"
  • "tf_flowers"

• structured

  • "higgs"
  • "iris"
  • "titanic"

• text

  • "cnn_dailymail"
  • "glue"
  • "imdb_reviews"
  • "lm1b"
  • "multi_nli"
  • "squad"
  • "wikipedia"
  • "xnli"

• translate

  • "flores"
  • "para_crawl"
  • "ted_hrlr_translate"
  • "ted_multi_translate"
  • "wmt14_translate"
  • "wmt15_translate"
  • "wmt16_translate"
  • "wmt17_translate"
  • "wmt18_translate"
  • "wmt19_translate"

• video

  • "bair_robot_pushing_small"
  • "moving_mnist"
  • "starcraft_video"
  • "ucf101"

在安装tensorflow-datasets时间，如果不是最新安装的环境，大概率会碰到以下问题：

1）tensorflow-datasets仅支持最新版，需要更新tensorflow，指令

pip3 install --upgrade tensorflow-gpu

2）'_NamespacePath' object has no attribute 'sort'问题，这个问题在于setuptool的问题，如果出现此问题，一般情况下还是版本兼容性问题，要做的就是升级setuptool，可以参考这篇博客：https://blog.csdn.net/Dickson531/article/details/78043627