Coding Projects and Duga Heaven :)

Duga

00:00

RELOAD!

[Hosch250/Rubberduck] 2 commits. 626 additions. 172 deletions.

[rubberduck-vba/Rubberduck] 2 commits. 16 issue comments. 2 additions. 2 deletions.

[skiwi2/OGamePlanner] 2 commits. 200 additions.

[Vogel612/TranslationHelper\‌] 1 issue comment.

Simon Forsberg

Okay, this seems to work better.

Most important part of the solution: Change number of backpropagation iterations from 1000 to 100 000

Tweeting Duck

Note to self: it's "ja", not "jp" - isn't this awesome? Rubberduck #localized in #Japanese! Yay #OSS! https://t.co/7keXei8rH3

Tweeted by rubberduckvba on February 6, 2016 at 11:58 PM

Simon Forsberg

I have wondered about this for hours and now it works!!

Self note: Implement automatic backpropagation termination so that I don't have to specify number of iterations.

Simon Forsberg

00:24

WinsLosses{wins=1701, losses=4, total=1705, percentage=0.9976539589442815}

It sure beats the crap out of a randomly playing AI.

Let's see if it can beat the crap out of me...

Simon Forsberg

00:38

Well... if I play my best, it has no chance of beating me.

Okay, now it just beat me two games in a row.

SirPython

What are you playing it in?

Simon Forsberg

00:55

"The 21 game". I figured that would be the easiest to start with.

Once I am happy with the performance here, I'll work on Tic-Tac-Toe.

When that's done, I'll try to expand it more and more to play more and more advanced games.

But I think the 21 game and regular TTT is enough of a challenge for now.

SirPython

Is the 21 game == blackjack?

Simon Forsberg

Nope

5

Q: Game counting to 21

The idea of the game is that the player and the computer take turns entering either 1, 2, or 3, and that number is added to a running total. Whoever goes over 21 loses. I need to develop a strategy that lets the computer always win. I have a strategy worked out, but I was wondering if there is a ...

5

Q: Monte Carlo AI in 21 game

I am very interested in the Monte Carlo AI. I tried my best, still, this AI plays very badly. This code "works" in the meaning that it does not crash, but the quality of play is extremely low. Have I completly misunderstood the Monte Carlo AI or is there just a nasty bug in my code preventing ...

python game ai

16

Q: 21: The number with the curse

The goal of this game is to not say the number 21. Rules: You must only say numbers consecutive to the previous said number. You may say 1, 2, or 3 numbers at a time. For example: Player 1 says 1 Player 2 says 2, 3 Player 1 says 4, 5, 6 Player 2 says 7, 8 And so on... You will be playi...

java ai

There's a whole bunch of implementations of it on Code Review.

Tweeting Duck

@Kotaro_of_GGG thank @yuominae on #github for that! We're still missing a bunch of newer keys, we'll need another PR.. wink-nudge ;-)

Tweeted by rubberduckvba on February 7, 2016 at 12:52 AM in reply to Kotaro_of_GGG

SirPython

02:01

Does anyone here use Atom?

Simon Forsberg

10:31

@SirPython What's Atom?

skiwi

12:27

hey

@SimonForsberg I believe it's a text editor

@SimonForsberg 1000 is indeed nothing

@SimonForsberg Wait, what is it doing again?

Simon Forsberg

12:41

@skiwi It's playing the 21 game against a randomly playing AI

Duga

[Zomis/Machine-Learning] Zomis opened issue #1: Use Matrixes for weights

> Use Apache Commons RealMatrix and keep each layer's weights as a Matrix and use that Matrix to calculate activations and stuff.

[Zomis/Machine-Learning] Zomis opened issue #2: Use Apache Commons optimizers

> Try using Apache Commons Optimizers instead of the current Backpropagation

[Zomis/Machine-Learning] Zomis opened issue #3: Try other Neural Network libraries

> Look into Neuroph and Deep Learning 4j and measure the performances of those.

Dan Pantry

@SirPython ah, okay, fair enough.

Duga

[Zomis/Machine-Learning] Zomis opened issue #4: Implement regularization for Neural Networks

> Add regularization for Neural Networks, in both cost function and gradient calculation of course.

Duga

13:08

[Zomis/Machine-Learning] Zomis deleted branch clustering

Marc-Andre

14:07

@SirPython I'm using it!

Phrancis

17:16

Hey

Simon Forsberg

17:32

Hi @Phrancis

I just implemented Stochastic Gradient Descent. Holy crap that was effective.

WinsLosses{wins=9817, losses=183, total=10000, percentage=0.9817}

I am not 100% satisfied, the AI is still able to lose every now and then, but I'd say this is a significant achievement.

This is the first step towards an AI that really can learn to play almost any type of game.

Incoming commits!

Duga

[Zomis/Machine-Learning] Zomis created branch develop

[Zomis/Machine-Learning] Zomis pushed commit b4175f74 to develop: Continue work on GameAI

[Zomis/Machine-Learning] Zomis pushed commit 1e103a16 to develop: Add feature extraction of Integer to GameAI, add some learning

[Zomis/Machine-Learning] Zomis pushed commit dce837c0 to develop: Add GameAI name

[Zomis/Machine-Learning] Zomis pushed commit 16bdb0fa to develop: Decrease backprop lograte in GameAI

[Zomis/Machine-Learning] Zomis pushed commit 983f02cf to develop: Use the trained Neural Network in GameAI

[Zomis/Machine-Learning] Zomis pushed commit 05379931 to develop: Add some GameAI logging

[Zomis/Machine-Learning] Zomis pushed commit 55975718 to develop: Significantly improve GameAI, especially by using more backpropagation iterations

[Zomis/Machine-Learning] Zomis pushed commit 65b30f66 to develop: Add the possibility to play Human vs. AI

[Zomis/Machine-Learning] Zomis pushed commit 00ed2423 to develop: Add methods in MyRobot for right click and middle click

[Zomis/Machine-Learning] Zomis pushed commit 5d57f9f6 to develop: Add another test to BasicPropagationTest, with different size

[Zomis/Machine-Learning] Zomis pushed commit f6602f1f to develop: Cleanup in Backpropagation

[Zomis/Machine-Learning] Zomis pushed commit 647ca919 to develop: Change backpropagation logging

[Zomis/Machine-Learning] Zomis pushed commit b15f873c to develop: Variable change

[Zomis/Machine-Learning] Zomis pushed commit 2591ece4 to develop: Add flexibility to weight initialization in backprop, made it possible to skip entirely

[Zomis/Machine-Learning] Zomis pushed commit 5ce28847 to develop: Make use of flexible way to initialize weights

[Zomis/Machine-Learning] Zomis pushed commit 21ee0103 to develop: Add another, unused, way of initializing weights which doesn't seem effective

[Zomis/Machine-Learning] Zomis pushed commit bea5c70a to develop: Add training data weights, not sure if it's a good idea yet

[Zomis/Machine-Learning] Zomis pushed commit 5ab9eefb to develop: Add Backpropagation.costFunctionCheck to make sure that cost function is decreasing

[Zomis/Machine-Learning] Zomis pushed commit 43055af8 to develop: Use stochastic gradient descent in GameAI

[Zomis/Machine-Learning] Zomis pushed commit 85327f29 to develop: Adjust logging and what features to use

[Zomis/Machine-Learning] build for commit 55975718 on develop: The Travis CI build failed

BUILD FAILURE!

[Zomis/Machine-Learning] build for commit b15f873c on develop: The Travis CI build failed

BUILD FAILURE!

[Zomis/Machine-Learning] build for commit 85327f29 on develop: The Travis CI build failed

BUILD FAILURE!

Simon Forsberg

17:52

Erhm... perhaps it's pretty stupid to have that really huge test in Test...

4

A: Android gradle build script returns error 137 in preDexDebug

The problem was due to memory problems on the build server. Switching to a build server with more memory has fixed the problem.

Duga

[Zomis/Machine-Learning] Zomis pushed commit d4ea2409 to develop: Add some webscraping methods for fetching from Stack Exchange chat

[Zomis/Machine-Learning] Zomis pushed commit d520acda to develop: WIP: Add some text analysis methods

[Zomis/Machine-Learning] Zomis pushed commit b7de58ed to develop: Don't use the full Game21 playing as a Test

skiwi

@SimonForsberg Holy commits

Duga

[Zomis/Machine-Learning] Zomis deleted branch text

[Zomis/Machine-Learning] Zomis deleted branch images

Simon Forsberg

@skiwi Normally @Duga only posts 10 commits when there are more than 10 being pushed, but I pushed a few commits at a time just so that she would post them all :)

Figured you might want to read the commit messages to see what I've been doing.

skiwi

You're using words I know

Duga

17:58

[Zomis/Machine-Learning] build for commit b7de58ed on develop: The Travis CI build passed

[Zomis/Machine-Learning] Zomis pushed 130 commits to master

[Zomis/Machine-Learning] build for commit b7de58ed on master: The Travis CI build passed

Simon Forsberg

19:43

Started by user Simon Forsberg
Building in workspace /var/lib/jenkins/workspace/deploy-cardshifter
Cloning the remote Git repository
Cloning repository github.com/Cardshifter/cardshifter.github.io
 > git init /var/lib/jenkins/workspace/deploy-cardshifter # timeout=10
Fetching upstream changes from github.com/Cardshifter/cardshifter.github.io
 > git --version # timeout=10
 > git -c core.askpass=true fetch --tags --progress github.com/Cardshifter/cardshifter.github.io +refs/heads/*:refs/remotes/origin/*

Something is happening but not what I was hoping for...

@Marc-Andre @Phrancis Can anyone of you fix this problem below?

Tests in error:
  testPlayCardshifter(com.cardshifter.io.web.controller.StaticControllerTest): Request processing failed; nested exception is org.thymeleaf.exceptions.TemplateInputException: Error resolving template "play-cardshifter", template might not exist or might not be accessible by any of the configured Template Resolvers (template:11)

Simon Forsberg

19:59

@Marc-Andre @Phrancis Can anyone of you confirm that cardshifter.com has been updated to the most recent version?

Jan 14 at 10:07, by Simon Forsberg

Remind me to consider using Jenkins for deployment of Cardshifter webpage, maybe even Cardshifter server, Cardshifter snapshot, Cardshifter release to github, and @Duga

Cardshifter webpage, check!

Cardshifter snapshot would be really easy to fix as well

Cardshifter release to github can also be handled

As for @Duga and Cardshifter server, that I am not so sure about what I was planning...

Or maybe @Duga as in downloading and deploying latest version of @Duga? That should be accomplishable.

I'd probably also have use for building most recent version of my Minesweeper server and other things

Although I rarely develop the Minesweeper server

Phrancis

Not at home right now, will check when I get home

Duga

20:19

[Zomis/Duga] Zomis closed issue #60: Out of memory - possible memory leak

Tweeting Duck

@Unihedro are we tomorrow yet? bwahahaha #GitHub #TomorrowWillBeGood https://t.co/ohuMJh27cC

Tweeted by rubberduckvba on February 7, 2016 at 8:26 PM

Duga

[Zomis/Duga] Zomis added label opinions-wanted to issue #83: Accept some commands from a chat room user

[Zomis/Duga] Zomis added label feature-request to issue #83: Accept some commands from a chat room user

[Zomis/Duga] Zomis added label chat-interactivity to issue #83: Accept some commands from a chat room user

[Zomis/Duga] Zomis reopened issue #83: Accept some commands from a chat room user

[Zomis/Duga] Zomis commented on issue #83: Accept some commands from a chat room user

> Re-opening this as I haven't totally rejected the possibility of this in the future.

[Zomis/Duga] Zomis commented on issue #2: The project should be split into subprojects

> Re-opening this as this project would benefit of having some parts of it extracted, such as the core chat interactivity (posting to chat and polling chat)

[Zomis/Duga] Zomis removed label wontfix from issue #2: The project should be split into subprojects

[Zomis/Duga] Zomis reopened issue #2: The project should be split into subprojects

Phrancis

21:39

Shopping for a new notebook I'm getting less and less interested in MacBook, seems like the value vs. cost is just not really there

SirPython

23:07

@SimonForsberg It's an IDE; haven't you seen it before on Github?

At one time not too long ago, Github posted an "ad" for it.

Duga

23:45

200_success vs. rolfl: 9318 diff. Year: +2482. Quarter: +2482. Month: +941. Week: +90. Day: +90.

Mat's Mug vs. Simon Forsberg: 3617 diff. Year: +331. Quarter: +331. Month: +331. Week: +40. Day: +40.

200_success vs. janos: 19476 diff. Year: +2706. Quarter: +2706. Month: +866. Week: +70. Day: +70.

Loki Astari vs. Simon Forsberg: 2552 diff. Year: -114. Quarter: -114. Month: -9. Week: +20. Day: +20.

Transcript for

Coding Projects and Duga Heaven :)