Tensorflow median value

问题:

How can I calculate the median value of a list in tensorflow? Like

node = tf.median(X)

X is the placeholder
In numpy, I can directly use np.median to get the median value. How can I use the numpy operation in tensorflow?

回答1:

edit: This answer is outdated, use Lucas Venezian Povoa's solution instead. It is simpler and faster.

You can calculate the median inside tensorflow using:

def get_median(v):
    v = tf.reshape(v, [-1])
    mid = v.get_shape()[0]//2 + 1
    return tf.nn.top_k(v, mid).values[-1]

If X is already a vector you can skip the reshaping.

If you care about the median value being the mean of the two middle elements for vectors of even size, you should use this instead:

def get_real_median(v):
    v = tf.reshape(v, [-1])
    l = v.get_shape()[0]
    mid = l//2 + 1
    val = tf.nn.top_k(v, mid).values
    if l % 2 == 1:
        return val[-1]
    else:
        return 0.5 * (val[-1] + val[-2])

回答2:

For calculating median of an array with tensorflow you can use quantile function, since 50% quantile is the median.

import tensorflow as tf
import numpy as np 

np.random.seed(0)   
x = np.random.normal(3.0, .1, 100)

median = tf.contrib.distributions.percentile(x, 50.0)

tf.Session().run(median)

This code has not the same behavior of np.median because interpolation parameter approximates the result to lower, higher or nearest sample value.

If you want same behavior you could use:

median = tf.contrib.distributions.percentile(x, 50.0, interpolation='lower')
median += tf.contrib.distributions.percentile(x, 50.0, interpolation='higher')
median /= 2.
tf.Session().run(median)

Besides that, code above is equivalent to np.percentile(x, 50, interpolation='midpoint').

回答3:

We can modify BlueSun's solution to be much faster on GPUs:

def get_median(v):
    v = tf.reshape(v, [-1])
    m = v.get_shape()[0]//2
    return tf.reduce_min(tf.nn.top_k(v, m, sorted=False).values)

This is as fast as (in my experience) using tf.contrib.distributions.percentile(v, 50.0), and returns one of the actual elements.

回答4:

There is currently no median function in TF. The only way to use numpy operation in TF is after you run your graph:

import tensorflow as tf
import numpy as np

a = tf.random_uniform(shape=(5, 5))

with tf.Session() as sess:
    np_matrix = sess.run(a)
    print np.median(np_matrix)