CODEWITHSUNDEEP
pythontensorflow
itsamineral
itsamineral

Reputation: 1409

tensorflow map_fun indexing

I am trying to use the tensorflow map function but I am stuck at an indexing problem.

In simple python, I am trying to do the following operation:-

for i in range(1,25):
      u [i] = uold [i] - K * ( uold [i] - uold [i-1] )

In tensorflow, I am encountering an indexing issue due to "(uold1[i]-uold1[i-1])". Currently I have written the statement as:-

 u = tf.map_fn ( lambda u: uold - K * ( uold - uold ), uold )

In the current equation the second term is always zero. I am not sure how to change it to get the desired output.

Upvotes: 0

Views: 91

Answers (1)

David Wong
David Wong

Reputation: 748

You probably want to create a tensor that shifts to right by one dimension (using tf.pad()) instead, and then calculate the difference. Ex.

temp = uold_shifted_to_right - K * (uold - uold_shifted_to_right)

Then take out the first column from temp (using tf.slice()).

Upvotes: 1

Related Questions