Re: Can we neural networks to predict code change?

If it can predict imminent *place* (forget content) of change >50% of
time, it could be tremendously helpful.

That means huge gains in being able to predict our release cycles and
marketing and even customer support.

All we need is an big enough, properly trained ANN that predicts how a
graph evolves from time to time.

Has anyone done this?

-Bhu***

Chris Smith wrote:
<joshipura@xxxxxxxxx> wrote:
I have never studied theoretical computer science. However, I know a
few facts:

2. Neural networks are not digital computers - and can solve somethings
Turing machine can not

If you mean a form of neural networks that can be implemented on any
ordinary computer, then you are wrong; neural networks can't do anything
that's not possible with a Turing machine. If nothing else, a Turing
machine could at least simulate the neural network performing the
computation.

If you are referring to biological neural networks -- e.g., human brains
-- and asserting that they are capable of things that couldn't be
simulated by a computer at all, then this is philosophy rather than
computer science.

My question:
Is it theoretically possible to
A. Train a neural network about past and current code in form of change
in graph and
B. Ask it to predict what the next change could be?

Our brain predicts which module might fail next while maintaining a
large code base - with reasonable accuracy. Why can't it be done
through neural networks?

As was already said, there are certainly going to be reasonable
heuristics, and artificial neural networks (and therefore Turing
machines) may be able to pick up on them and pick them out of a body of
past experience and application of logic. I would conjecture that this
is essentially what the human brain does, as well. However, it's clear
that neither the human brain nor any kind of computer can give an exact
answer in the general case.

--
Chris Smith

.

Quantcast