使用标签谓词的tf.data过滤数据集

Question

我正在尝试使用下面给出的特定标签过滤CIFAR10培训和测试数据，

import tensorflow as tf
from tensorflow.keras import datasets, layers, models
import tensorflow_datasets as tfds
import matplotlib.pyplot as plt
import numpy as np

数据集

dataset = datasets.cifar10.load_data()

分割数据集

train_data = tf.data.Dataset.from_tensor_slices((dataset[0][0],dataset[0][1]))
test_data = tf.data.Dataset.from_tensor_slices((dataset[1][0],dataset[1][1]))

过滤功能

def filter_f(datas,filter_labels = tf.constant([0,1,2])):
  x = tf.not_equal(datas[1],filter_labels)
  x = tf.reduce_sum(tf.cast(x, tf.uint8))
  return tf.greater(x, tf.constant(0,tf.uint8))

dataset = train_data.filter(filter_f).batch(200)

根据similar issue。但是，filter函数返回上面代码中未过滤的内容。

labels = []
for i, x in enumerate(tfds.as_numpy(dataset)):
    labels.append(x[1][0][0])
print(labels)

返回

[4, 7, 5, 6, 0, 5, 5, 6, 5, 3, 6, 7, 0, 0, 6, 3]

要再现结果，请使用此colab link

Answer 1