I would like to do something like:
x = tf.Variable(tf.ones([100], dtype=tf.float32))
x0 = tf.Variable(tf.ones([1], dtype=tf.float32))
def f(x):
return tf.sin(x)
x[0] = x0
for i in range(1,100):
x[i+1] = f(x[i])
to construct a tensor 'x'. Is such a construct possible? I looked at 'tf.while_loop' but does not seem to help.
def f(x):
return tf.sin(x)
x = []
x.append(tf.constant(1.0))
for i in range(1,100):
x.append(f(i))
Is this what you want? This is not recursive by the way. It's iterative.
The solution in TypeError: 'Tensor' object does not support item assignment in TensorFlow pretty much solves the problem as Atirag pointed out, but 'tf.pack()' seems to have been deprecated. Just for completeness, here is an up-to-date solution to the posted problem:
x0 = tf.Variable(tf.ones(1, dtype=tf.float32))
def f(x):
return tf.sin(x)
x = [x0]
for i in range(1, 100):
x.append(f(x[i - 1]))
tf.stack(x)
