tensorflow计算模型分类准确率accuracy的几种方法讲解

记录一下分别用：tf.equal()；tf.nn.in_top_k()；tf.metrics.accuracy() 计算分类准确率accuracy的过程。

1. 使用tf.equal()函数：

* 要求参数x和y有相同的类型

import tensorflow as tf

batch_size = 8
class_num = 10  # 类别数量

# 定义一个logits为神经网络预测的标签结果，shape:(batch_size,)
logits = tf.constant([0, 5, 9, 1, 7, 1, 0, 1])

# 定义一个labels为真实样本号，这里设为全1，shape:(batch_size,)
labels = tf.ones((batch_size,), dtype=tf.int32)

# 使用tf.equal()计算分类准确率，返回的是bool类型的Tensor来判断对应位的值是否相等
correct_prediction = tf.equal(logits, labels)
# 将bool值转为float类型（即0或1）再求均值即为准确率
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))

with tf.Session() as sess:
    print(logits.eval())
    print(labels.eval())
    print(correct_prediction.eval())
    print("accuracy:{}".format(accuracy.eval()))

打印结果为:

[0 5 9 1 7 1 0 1]
[1 1 1 1 1 1 1 1]
[False False False  True False  True False  True]
accuracy:0.375

 

2. 使用tf.nn.in_top_k()函数：

* 要求参数'predictions'为float32类型，且shape为(batch_size， class_num)

* 要求参数'targets'为int32或int64类型，且shape为(batch_size，)

import tensorflow as tf

batch_size = 8
class_num = 10  # 类别数量

# 定义一个logits为神经网络预测的标签结果，shape:(batch_size, class_num)
logits = tf.constant([0, 5, 9, 1, 7, 1, 0, 1])
# 通过one_hot函数将(batch_size, )转为(batch_size, class_num)
logits = tf.one_hot(indices=tf.cast(logits, tf.int32), depth=class_num)

# 定义一个labels为真实样本号，这里设为全1，shape:(batch_size,)
labels = tf.ones((batch_size, ), dtype=tf.int32)

# 使用tf.nn.in_top_k()计算分类准确率，当参数k=1时等价于tf.equal()
# 返回的是bool类型的Tensor来判断对应位的值是否相等
correct_prediction = tf.nn.in_top_k(logits, labels, 1)
# 将bool值转为float类型（即0或1）再求均值即为准确率
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))

with tf.Session() as sess:
    print(logits.eval())
    print(labels.eval())
    print(correct_prediction.eval())
    print("accuracy:{}".format(accuracy.eval()))

打印结果为：

[[1. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
 [0. 0. 0. 0. 0. 1. 0. 0. 0. 0.]
 [0. 0. 0. 0. 0. 0. 0. 0. 0. 1.]
 [0. 1. 0. 0. 0. 0. 0. 0. 0. 0.]
 [0. 0. 0. 0. 0. 0. 0. 1. 0. 0.]
 [0. 1. 0. 0. 0. 0. 0. 0. 0. 0.]
 [1. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
 [0. 1. 0. 0. 0. 0. 0. 0. 0. 0.]]
[1 1 1 1 1 1 1 1]
[False False False  True False  True False  True]
accuracy:0.375

 

3.使用tf.metrics.accuracy()函数：

* 要求参数labels要和参数predictions的shape相同

import tensorflow as tf

batch_size = 8
class_num = 10  # 类别数量

# 定义一个logits为神经网络预测的标签结果，shape:(batch_size, )
logits = tf.constant([0, 5, 9, 1, 7, 1, 0, 1])
# 定义一个labels为真实样本号，这里设为全1，shape:(batch_size, )
labels = tf.ones((batch_size, ), dtype=tf.int32)

# 使用tf.metrics.accuracy()计算分类准确率，返回的第一个值即为分类准确率
acc, acc_op = tf.metrics.accuracy(logits, labels)

with tf.Session() as sess:
    sess.run(tf.local_variables_initializer())
    print(logits.eval())
    print(labels.eval())
    print("accuracy:{}".format(acc_op.eval()))

打印结果为：

[0 5 9 1 7 1 0 1]
[1 1 1 1 1 1 1 1]
accuracy:0.375