Closed StrangeTcy closed 7 years ago
I don't think this code is actively maintained or used anywhere. You may want to contact Ted Sanders @tpsjr7 directly to see what the deal is.
Perhaps, and yet
Hello,
I've moved on from this project due to my personal situation and lack of free time. I would consider this particular implementation abandoned.
You may want to post a follow up to this thread
https://groups.google.com/d/msg/opencog/l0Bp5zkimJo/HiEfs9j_AgAJ
On Mon, Aug 29, 2016 at 2:08 AM, StrangeTcy notifications@github.com wrote:
Perhaps, and yet
- This problem could still be solvable
- OpenCog intends to integrate DeSTIN as far as I know, so shouldn't OpenCog people try to give this whole repository some attention?
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/opencog/destin/issues/20#issuecomment-243048156, or mute the thread https://github.com/notifications/unsubscribe-auth/ACcjNUu6U-2n8MhoHiuJ3ziSqESa7ngOks5qkoVsgaJpZM4JvHoy .
Oh well. But at least that's some clarity. Thanks for the heads-up, Ted :-)
Also, does that make the issue closed?
Well, if you think that you have some spare time, and were interested in taking over and working on this project, then you should contact Ben Goertzel to discuss. Otherwise, this is an issue that ... will get rotten and moldy with time.
I should probably finish reading EGi first.
Hi. I played with DeSTIN long time ago. Now I've checked it using fresh Ubuntu 16.04 with SWIG 2.0.12 installed from sources and it still works for me! idle som.py
after some training iterations displays a window with map visualization. Are you sure you followed the build instruction precisely?
Regarding plans, I suppose OpenCog is going to integrate TensorFlow or Theano rather than DeSTIN, but I don't know if someone is working on that.
Hi Jacek,
Ben did mumble something about using DeSTIN ideas to provide an overall control-framework for tensorflow or theano, or something like that, but I don't understand any details. I think he was less than happy about how well DeSTIN worked, but still felt that the raw theory of it was good and salvageable.
I am, but double-checking wouldn't hurt. I might report the results once I try to solve this issue myself.
As for Tensorflow or Theano, my general impression was that a machine using OpenCog should eventually become capable of "gobbling up" and integrating any neural network written in any language. And regarding working on it -- I might, but much later on, probably.
Hi @StrangeTcy. I have created Dockerfile (https://github.com/opencog/destin/blob/master/Dockerfile), maybe this will help you to install dependencies.
Thanks!
I've built it, but in there trying idles som.py
leads to
Traceback (most recent call last):
File "/usr/bin/idle", line 5, in
main() File "/usr/lib/python2.7/idlelib/PyShell.py", line 1553, in main root = Tk(className="Idle") File "/usr/lib/python2.7/lib-tk/Tkinter.py", line 1818, in init self.tk = _tkinter.create(screenName, baseName, className, interactive, wantobjects, useTk, sync, use) _tkinter.TclError: no display name and no $DISPLAY environment variable
I tried export DISPLAY=0.0
, but then idle som.py
said it
_tkinter.TclError: couldn't connect to display "localhost:0.0"
Trying docker run -it --env="DISPLAY" opencog/destin
didn't help either.
This line works for me on Ubuntu:
docker run -v /tmp/.X11-unix:/tmp/.X11-unix -e DISPLAY=:0 -it opencog/destin
That helped. thanks a lot.
Another issue: for all the cool things that are installed when this Dockerfile is used, I still have to install idle
manually.
Yet another issue: I cannot use vi properly, and i wished to use an editor with a gui to edit som.py
and point it to the right location of the cifar-10-batches-bin
directory.
But gedit
errored on me:
No protocol specified Failed to connect to Mir: Failed to connect to server socket: No such file or directory Unable to init server: Could not connect: Connection refused
(gedit:5563): Gtk-WARNING **: cannot open display: :0
I had to do some cp -R
to copy the cifar files to the location that was in the som.py
originally.
And then it worked like a charm.
idle dostuff.py
worked as well, if slowly.
Is it supposed to output a lot of calculations to console when running?
I have updated Dockerfile in #23, so it downloads CIPHAR data into proper directory as well as imports idle packages. It is not so easy to use gedit
from inside docker container. But you can use idle
instead. This is a development environment for python, seems to suffice at least for small changes.
dostuff.py
works for me for about 30 seconds in total. It displays belief states for each layers nodes. After some training time the belief of 7th layer's single node corresponds with the state of the hand (open, closed, changing). This shows that the network learns. The console is supposed to display current belief states, so it outputs a lot.
Thanks!
Oh. I didn't realise that about idle
.
For me, running
idle som.py
on Ubuntu 16.04 x64 prodcues a window, as expected. Pressing<F5>
in this window produces another window, which outputs an error message: