Fix tf.nn.embedding_lookup() InvalidArgumentError: indices = 28 is not in [0, 28) – TensorFlow Tutorial

We often use tf.nn.embedding_lookup() to pick up elements in a tensor by ids. Here is an tutorial.

Understand tf.nn.embedding_lookup(): Pick Up Elements by Ids – TensorFlow Tutorial

However, it may encounter InvalidArgumentError: indices is not error when using this function. In this tutorial, we will introduce how to fix this error.

Please look at example code below:

import tensorflow as tf
import numpy as np

z = tf.Variable(np.array([[2,2,3],[6,7,2]], dtype = np.float32), name = 'x1')
xw = tf.nn.embedding_lookup(z,4)

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    print(sess.run(xw))

Run this python code, you will find InvalidArgumentError: indices is not error occur.

Why does this InvalidArgumentError occur?

The reason is the id is out of tensor induce. As to example above, there are only 2 elements on axis = 0, however, the id is 4, which is out of [0, 2)

How to fix this InvalidArgumentError?

Making the id is in [0,2)

For example:

xw = tf.nn.embedding_lookup(z,1)

Then you will find this InvalidArgumentError is fixed.