tf.nn.top_k() tf.nn.in_top_k()

1..中文

tf.nn.top_k(input, k, name=None)

解释：这个函数的作用是返回 input 中每行最大的 k 个数，并且返回它们所在位置的索引。

#!/usr/bin/env python
# -*- coding: utf-8 -*-
import tensorflow as tf
import numpy as np

input = tf.constant(np.random.rand(3,4))
k = 2
output = tf.nn.top_k(input, k)
with tf.Session() as sess:
print(sess.run(input))
print(sess.run(output))

[[ 0.98925872  0.15743092  0.76471106  0.5949957 ]
[ 0.95766488  0.67846336  0.21058844  0.2644312 ]
[ 0.65531991  0.61445187  0.65372938  0.88111084]]
TopKV2(values=array([[ 0.98925872,  0.76471106],
[ 0.95766488,  0.67846336],
[ 0.88111084,  0.65531991]]), indices=array([[0, 2],
[0, 1],
[3, 0]]))

输入参数：
input: 一个张量，数据类型必须是以下之一：float32、float64、int32、int64、uint8、int16、int8。数据维度是 batch_size 乘上 x 个类别。
k: 一个整型，必须 >= 1。在每行中，查找最大的 k 个值。
name: 为这个操作取个名字。

输出参数：

一个元组 Tensor ，数据元素是 (values, indices)，具体如下：

values: 一个张量，数据类型和 input 相同。数据维度是 batch_size 乘上 k 个最大值。

indices: 一个张量，数据类型是 int32 。每个最大值在 input 中的索引位置。

tf.nn.in_top_k(predictions, targets, k, name=None)

解释：这个函数的作用是返回一个布尔向量，说明目标值是否存在于预测值之中。

输出数据是一个 targets 长度的布尔向量，如果目标值存在于预测值之中，那么 out[i] = true。

注意：targets 是predictions中的索引位，并不是 predictions 中具体的值。

#!/usr/bin/env python
# -*- coding: utf-8 -*-
import tensorflow as tf
import numpy as np

input = tf.constant(np.random.rand(3,4), tf.float32)
k = 2   #targets对应的索引是否在最大的前k(2)个数据中
output = tf.nn.in_top_k(input, [3,3,3], k)
with tf.Session() as sess:
print(sess.run(input))
print(sess.run(output))

[[ 0.43401602  0.29302254  0.40603295  0.21894781]
[ 0.77089119  0.95353228  0.04788217  0.37489092]
[ 0.83710146  0.2505011   0.28791779  0.97788286]]
[False False  True]

http://www.jianshu.com/p/343c2eaacd18
2. 英文：

tf.nn.top_k(input, k, name=None)

Returns the values and indices of the k largest elements for each row.


 represents the j-th largest element in 

.


 gives the column index of the corresponding element,such that 

.
If twoelements are equal, the lower-index element appears first.

Args:

input

: A 

Tensor

. Must be one of the following types: 

float32

, 

float64

, 

int32

, 

int64

, 

uint8

, 

int16

, 

int8

. A batch_size x
classes tensor

k

: An 

int

 that is 

>= 1

. Number of top elements to look for within each row

name

: A name for the operation (optional).

Returns:

A tuple of 

Tensor

 objects (values, indices).

values

: A 

Tensor

. Has the same type as 

input

. A batch_size x k tensor with the k largest elements for each row, sorted in descending order

indices

: A 

Tensor

 of type 

int32

. A batch_size x k tensor with the index of each value within each row

tf.nn.in_top_k(predictions, targets, k, name=None)

Says whether the targets are in the top K predictions.

This outputs a batch_size bool array, an entry out[i] is true if theprediction for the target class is among the top k predictions amongall predictions for example i. Note that the behavior of InTopK differsfrom the TopK op in its handling of ties; if multiple
classes have thesame prediction value and straddle the top-k boundary, all of thoseclasses are considered to be in the top k.

More formally, let


 be the predictions for all classes for example i, 

 be
the target class for example i, 

 be the output for example i,




Args:

predictions

: A 

Tensor

 of type 

float32

. A batch_size x classes tensor

targets

: A 

Tensor

 of type 

int32

. A batch_size vector of class ids

k

: An 

int

. Number of top elements to look at for computing precision

name

: A name for the operation (optional).

Returns:

A 

Tensor

 of type 

bool

. Computed Precision at k as a bool Tensor