通过正则表达式模式将TF.string分隔为定界符

Question

I want to split string by regex pattern in Tensorflow(TF). I found tf.string_split()

But, it expects delimiter as string not regex. How to do that in TF 1.2.

Code Sample :

data = tf.Variable("aa the boy aaa the boy aaaaa")
data_split = tf.string_split(data, re.compile("a*"))
TypeError: Expected string, got re.compile('.') of type 'SRE_Pattern' instead.

Answer 1

As of august 2017 there is no op in tensorflow to do this, but it's straightforward to implement one by copying the C++ code in StringSplit and editing it. Alternatively you can use tf.py_func to split in python instead of tensorflow.