Tensorflow-Hacks

usage:

• import immutable_scatter_nd_update from scatter_nd_for_immutable_tensor.py
• Just invoke immutable_scatter_nd_update(args) instead of usual tensorflow's scatter_nd_update(args) in your code.
• All arguments are same as scatter_nd_update(args) provided by tensorflow

examples with usage

example 1

tf.InteractiveSession()
inp1 = tf.constant(np.ones([3,4,5,2,2]),tf.int32)
inp2 = tf.constant([[0,0,0],[1,1,1],[0,1,1]])
inp3 = tf.constant([[111,111],[222,222],[4,5]])
inp3 = tf.stack(2*[inp3],axis=2)
x = immutable_scatter_nd_update(inp1,inp2,inp3)
print x[0,0,0].eval()
print x[1,1,1].eval()
print x[0,1,1].eval()

example 2

inp1 = tf.constant(np.ones([3,4,5]),tf.int32)
inp2 = tf.constant([[0,0,0],[1,1,1],[0,1,1]])
inp3 = tf.constant([0,2,3])
x = immutable_scatter_nd_update(inp1,inp2,inp3)
print x[0,0,0].eval()
print x[1,1,1].eval()
print x[0,1,1].eval()