Tensorflow: Finding index of first occurrence of elements in a tensor

Question

Suppose I have a tensor, x = [1, 2, 6, 6, 4, 2, 3, 2]
I want to find the index of the first occurrence of every unique element in x.
The output should be [0, 1, 6, 4, 2]. I basically want the second output of numpy.unique(x,return_index=True). This functionality doesn't seem to be supported in tf.unique. Is there a workaround to this in tensorflow, without using any loops?

Answer 1

One possibilty would be to use x.eval() which gives you a numpy array back, then use numpy.unique(...)

Answer 2

x = [1, 2, 6, 6, 4, 2, 3, 2]
x_count = tf.cumsum(tf.ones_like(x))-1
unique, unique_id = tf.unique(x)
unique_first = tf.unsorted_segment_min(x_count, unique_id, tf.shape(unique)[0])
with tf.Session() as sess:
    print(sess.run(tf.stack([unique, unique_first],0)))

礼物：

[[1 2 6 4 3]
 [0 1 2 4 6]]