Coding Projects and Duga Heaven :) - 2016-03-01

« first day (577 days earlier) ← previous day next day → last day (2990 days later) »

12:00 AM

RELOAD!

[Hosch250/Rubberduck] 26 commits. 5731 additions. 2414 deletions.

[retailcoder/Rubberduck] 2 commits. 44 additions. 79 deletions.

[rubberduck-vba/Rubberduck] 2 commits. 3 opened issues. 1 closed issue. 18 issue comments. 64 additions. 48 deletions.

2 hours later…

2:00 AM

@Phrancis Out of curiosity, where are you learning about these data types?

I realize I need to learn some more about these things, along with some algorithms.

@SirPython amazon.com/Java-Beginners-Guide-Sixth-Edition/dp/0071809252

Also, just posted a Q on that:

0

Q: FiniteArrayQueue type with interface and unit tests

Doing some exercises on basic data structures, I learned about queues, and decided to roll my own to better understand how a basic queue can function. As the name indicates, this Queue is made from a finite (i.e. regular) array. The JavaDoc gives more details throughout. All this is pretty new ...

java array queue interface junit

6 hours later…

8:10 AM

@SirPython Review old JDQ's.

3.8k 12 49 97

java algorithm data-structures tree linked-list matrix array

2 hours later…

10:38 AM

Afternoon all

hey

/s/Afternoon/Morning/

1 hour later…

12:04 PM

Hey

Man writing unit tests is pretty time-consuming, it seems like I've been writing tests for that one class for like 2 hours and still not quite done

2 hours later…

2:24 PM

@Phrancis Yeah, but debugging takes even longer.

I must say, JUnit is pretty good, but I like Groovy's JUnit language & implementation better, much more succinct, and gives better assertion messages

1 hour later…

3:50 PM

@PinCrash Nothing beats Groovy's power-assertions.

PinCrash? Wtf?

lol... celebrating PPCG graduation announcement with the anagram name tradition :)

Oh :)

@SimonForsberg if you have anything to add on the above question I posted last night I'm all ears. Got some pretty good answers, especially the one about unit tests... man writing tests is alot more work than I initially thought :)

I might have a thing or two to say. Not sure if it's already been said though. Might take a look later.

4:07 PM

@PinCrash Why the sudden name change?

17 mins ago, by Pin Crash

lol... celebrating PPCG graduation announcement with the anagram name tradition :)

Argh I read over that one.

What does PPCG stand for?

115

Q: Congratulations, you're graduating!

It's a big day, and I know you folks have been waiting a reeeaaal long while. Double waiting, you might even say, since unfortunately we made you wait even longer after initial news. But, let's make it official! You've been cleared for graduation by the Stack Exchange Community Team! Programming ...

discussion featured graduation

Aha, I thought it was just called code golf

1 hour later…

5:15 PM

@SimonForsberg Any chance you looked into the Test/Train splitting? :)

gonna do that now, @DJanssens

Keep me updated.

You told me to take a look at @rolfl's benchmarking tool. However is it available in gradle. I noticed a Maven possibility. But never worked with Gradle before. Does a website exist that holds all existing libs?

@DJanssens Gradle and Maven are just two different build tools, both of them can work with dependencies.

<dependency>
    <groupId>net.tuis.ubench</groupId>
    <artifactId>ubench</artifactId>
    <version>0.1.0</version>
</dependency>

that's Maven style ^^

I'm aware of that. However for Maven there is "Maven central"

Gradle can take stuff from Maven central as well

5:25 PM

Hmm let's see.

mvnrepository.com/artifact/net.tuis.ubench/ubench/0.2.0

in Gradle you just add "compile 'net.tuis.ubench:ubench:0.2.0'" to the dependencies

You are right. I wasn't aware that maven central also offered Gradle. Thanks ;)

What are we aiming to do with the partitioning of the trainingset, @DJanssens? Analyze the F1 score? Make training graphs for analyzing bias/variance? Calculate the cost function for the different sets?

Hmm. I'm not 100% sure again what bias/variance was ;)

I guess analyzing the F1 score/precision and recall in a real setting.

Since now we have like 70-80% precision and recall. But I'm pretty sure that the model will perform poor in a real setting since it's overfitted.

bias = underfitting
variance = overfitting

5:34 PM

I have to reread that chapter :)

Anyway, normally the performance will drop with the test/train set. Then we can see how regularization contributes :)

[Zomis/Machine-Learning] Zomis opened issue #8: Don't use both TrainingData and LearningData

> Having both a TrainingData class and a LearningData class is confusing. Get rid of TrainingData, as that is only used for the gameai package.

@DJanssens good point

[Zomis/Machine-Learning] Zomis added label enhancement to issue #8: Don't use both TrainingData and LearningData

5:57 PM

[Zomis/Machine-Learning] Zomis pushed commit 6dd975f0 to develop: Add initial support for separating data into trainingset, cross-validationset and testset

How's that for a start, @DJanssens? ^^

Seems good to me. Except does it use a seed?

I'll see if I can apply this on the programmers comments

It would be nice if iterative runs end up with the same datasets for debugging purposes.

@DJanssens to perform the shuffling, yes it does. Pass it a random initialized with a seed and it will always generate the same values.

I tend to use new Random(42) in oh so many of my programs.

...and then I tend to forget to remove that 42 part later on...

6:01 PM

:D

[Zomis/Machine-Learning] build for commit 6dd975f0 on develop: The Travis CI build passed

hmm... I think I will make a slight API change

[Zomis/Machine-Learning] Zomis pushed commit a83abe7e to develop: Make PartionedDataSet store LearningDataSets to make use of the LearningDataSet APIs

Does that make sense, @DJanssens? ^^

[Zomis/Machine-Learning] Zomis pushed commit f1270094 to develop: Make use of PartitionedDataSet in ProgrammersCommentTest

6:09 PM

I was not aware this was possible: this(new ArrayList<>());

But I'm not aware of a lot of things.

Constructor-chaining. Really useful.

That calls the other constructor, right? But I thought it was required that the type was specified between <>

[Zomis/Machine-Learning] build for commit a83abe7e on develop: The Travis CI build passed

Why split the dataset in training+test+cross validation?

[Zomis/Machine-Learning] build for commit f1270094 on develop: The Travis CI build passed

[Zomis/Machine-Learning] Zomis pushed commit 0b10a2e6 to develop: Make more use of the partitioned data in ProgrammersCommentTest

6:14 PM

If I remember correctly cross validation is a technique that just runs the program multiple times. Each time taking a different 10% as test and 90% training data from the entire dataset. Right?

@DJanssens <> has been possible since Java 1.7. Welcome to the 21st century.

I've never used Java in depth ;) But I'm currently reading 'Java in a nutshell' to refine my knowledge ;)

@DJanssens that's not what I've been taught. Cross validation is for deciding the value of things like the lambda parameter. The value of lambda that gives the lowest cost function value on the crossvalid set is the value of lambda that will be used, and then you run it on the testset.

I'm currently running performance tests, but goes slow. I'll let you know the outcome @SimonForsberg

Isn't that just called the validation set?

Training Set Cost: 0.278353056693585
CrossValidation Cost: 0.30340004050836233
ALL Score: truePositive: 20, falsePositive: 5, trueNegative: 451, falseNegative: 50, precision 0.800000, recall 0.285714, f1 score: 0.421053
Training Score: truePositive: 20, falsePositive: 5, trueNegative: 451, falseNegative: 50, precision 0.800000, recall 0.285714, f1 score: 0.421053
CrossVal Score: truePositive: 3, falsePositive: 2, trueNegative: 90, falseNegative: 10, precision 0.600000, recall 0.230769, f1 score: 0.333333

that's what I get ^^

horrifyingly low f1 scores.

6:18 PM

[Zomis/Machine-Learning] build for commit 0b10a2e6 on develop: The Travis CI build passed

but something is wrong

ALL and Training should not be the same

        System.out.println("ALL Score: " + data.precisionRecallF1(learnedTheta, function));
        System.out.println("Training Score: " + data.precisionRecallF1(learnedTheta, function));

no wonder...

:D

Seems I was talking about K-fold cross validation

[Zomis/Machine-Learning] Zomis pushed commit 1f0b7088 to develop: Use training set for training score.

K-fold cross validation? Not sure what that is. Or do you mean the K-means clustering algorithm?

robjhyndman.com/hyndsight/crossvalidation

6:22 PM

Data is:
Training Set Cost: 0.278353056693585
CrossValidation Cost: 0.30340004050836233
ALL Score: truePositive: 20, falsePositive: 5, trueNegative: 451, falseNegative: 50, precision 0.800000, recall 0.285714, f1 score: 0.421053
Training Score: truePositive: 14, falsePositive: 3, trueNegative: 272, falseNegative: 26, precision 0.823529, recall 0.350000, f1 score: 0.491228
CrossVal Score: truePositive: 3, falsePositive: 2, trueNegative: 90, falseNegative: 10, precision 0.600000, recall 0.230769, f1 score: 0.333333

Hmm

I'm wondering why the F1 score for the entire training set dropped from 70+ to 42

I think my implementation and yours might differ a bit

[Zomis/Machine-Learning] build for commit 1f0b7088 on develop: The Travis CI build passed

are you using the automatically generated features?

Yes

The 50 most occurring ones

6:25 PM

How much data do you have? I added 250 training examples a week or so ago

Normally I used the version you pushed back then.

I changed my tests to be up to date with your version.

Not sure when or why it dropped this much

Anyway, when my performance test are finished I'll just pull your latest changes and see for myself ;)

Me neither :)

could do a git bisect probably to figure it out

seems to be because of a very low recall

which is because of quite many falseNegative

I think it changed when I added the 250 training examples

I set the press method in rolfl's benchmark tool to 100. Bad decision. Seems it takes quite some time to test it 100x.

6:30 PM

probably does

http://www.commitstrip.com/en/2016/03/01/the-end-of-the-coders/
CommitStrip
The end of the coders
CommitStrip
1456857195

2

@StackExchange Been thinking about that more than once.

Although it doesn't only apply to that. It also applies to general applications.

A programmers job is to make himself not needed.

True. Still I doubt that time will come anywhere soon.

I've seen some interesting applications though of Model driven development, where a program is written by interconnecting models (components).

Sounds interesting.

I honestly think that the time when AIs will code everything will never come. From studying Machine Learning, I understand better how much work actually has to be done to make a Machine Learn...

6:53 PM

True. Even google autocorrect messes up almost 1 word i every sentence I type.

Still a long way to go ;)

@SimonForsberg Do you use Java in your job? Or what kind of job do you have?

7:47 PM

@DJanssens For some values of Java. Java 6.

I'm currently unsure if I should enter the Java or .Net corporate world. :/ Since it is my first job I believe I can apply for both, nonetheless afterwards it will become much harder to switch I guess.

Java, Java, Java, Java, Java, Java, Java, Java, Java, Java, Java, Java

8:12 PM

Care to give a reason for your opinion?

I never used C# before, so I have the advantage that I know the IDE/Maven/hibernate in Java.

But I feel like its so bloated.

Tbh I love programming in Python. Sadly there are almost no job openings for that here in Belgium.

@DJanssens When it comes down to it, Java and C# are not that different, I know programmers at my work who have come in with Java background but adapted to C#, and vice-versa

I totally agree. Language wise its no biggie

However I believe its the frameworks

Like JSP/applets etc

@PinCrash I wanted to ask you, do you have a blog?

8:32 PM

@DJanssens I do not, at the moment

Too bad ;)

8:46 PM

@SimonForsberg Argh, Still processing the benchmark. And it already spend too much time to abort it now :D

9:31 PM

Monking.

°/ @SirPython

º/ @DJanssens! How're you doing?

Fine fine. looking for a job and trying to refresh my Java knowledge

What are you up to lately @SirPython

Nothing much. I've been pretty busy lately and am just starting to get back to React/ML.

Are you looking for Java or C# job? (I saw you discussing that earlier and I was wondering if you got any insight on a choice).

Just got some training in a functional area I know nothing at all about... *eyes glazed over*

9:38 PM

I do not. I'm not sure yet what I want myself. To be honest I haven't used Windows in the last 6 years. So I'm not that eager to switch to the .Net realm.

@PinCrash Functional area ==relatedto== functional programming?

@SirPython No, it's a "business area", or set of skills related to a particular area/field

@DJanssens I'd say check out old SO/other site questions, but all they do is start holy wars.

@PinCrash I see. That makes more sense

e.g., I'm specialized with EDIs/vendor interoperability, but this training was about patient care and clinical documentation

So, I was very much lost. Glad some other team members understand it better

@SirPython I did and indeed it's worthless. Both seem very much alike. I also checked the job opportunities in both directions (at my area) and they are similar.

Today I submitted my first job application. Huray!

3

9:41 PM

Great!

Congrats!

It's a longshot, since they searched for experienced developers.

9:58 PM

@DJanssens Java 8, Java 8, Java 8, Java 8, Java 8, Java 8, Java 8

Groovy and Scala are two good reasons to stay on the Java platform

Never really looked into those, which is a shame.

@DJanssens I find that there are more Java libraries than .net ones. (Maybe because of what I have been looking for, of course)

offtopic, how long does the commentTest take on your machine @SimonForsberg?

Well I never even heard of Machine learning libraries on .Net

@DJanssens Alright. Maybe I can take a look at your code for that another day, I should be familiar with rolfl's ubench library as I have reviewed it and contributed to it myself

I think it's normal. If I remember correctly your implementation takes around 3 minutes/run on my machine

10:01 PM

@DJanssens Well there's another reason to use Java :)

100 iterations is around 300 minutes => 5h +/-

@DJanssens Best of luck to you. I have really not sent in enough of those in my life. At least not in my programming life.

Enough of what?

@DJanssens Less than a minute I think. And that's with my old non-vectorized code...

That's odd.

But I'm using an old macbook '09.

10:03 PM

@DJanssens enough of job applications. I got the second job I sent a real application for.

Haha, I'm hoping for the same.

@DJanssens I'm using a somewhat new MSI laptop, running Windows

Windows :/

;)

@SimonForsberg You used Matlab for the ML Coursera course, right?

Octave I guess

At least, I did.

10:06 PM

Hm. The course said that Octave for Windows would not work.

Iirc you need to submit your exercises in octave

Hmm, I believe I used octave on windows.

I even have octave installed on my mobile

just for fun though :D

> Octave on Windows cannot be used to submit programming assignments in this course. The official installers do not support HTTPS, which we use to secure your submissions. We are working with the Octave team to resolve this issue. In the meantime, please use MATLAB.

Forgive me, I must have used ubuntu then.

@SirPython Nope, Octave.

@SirPython It works fine, you just need some patches.

Night

Night.

10:13 PM

Night @SimonForsberg

10:57 PM

Just found out... out of 2,274 repositories mentioning #VBA on #GitHub, @rubberduckvba ranks 2nd! Thanks stargazers! https://github.com/search?o=desc&q=vba&ref=searchresults&s=stars&type=Repositories&utf8=%E2%9C%93

Tweeted by rubberduckvba on March 1, 2016 at 10:55 PM

11:33 PM

Remonking.

Remonking to you

Hey @PinCrash! How're you doing?

Great! and you?

user image

About to give Clojure a spin^ for fun :)

Doing well, thanks!

Did that recent clojure question inspire you?

Um haven't seen it

Oh, there it is

Oh wow that next-apendage function is something else

11:45 PM

200_success vs. rolfl: 11015 diff. Year: +4179. Quarter: +4179. Month: +82. Week: +260. Day: +82.

Mat's Mug vs. Simon Forsberg: 3968 diff. Year: +682. Quarter: +682. Month: +40. Week: -194. Day: +40.

Loki Astari vs. Simon Forsberg: 2980 diff. Year: +314. Quarter: +314. Month: -15. Week: +13. Day: -15.

200_success vs. janos: 20194 diff. Year: +3424. Quarter: +3424. Month: +102. Week: +190. Day: +102.

« first day (577 days earlier) ← previous day next day → last day (2990 days later) »

Transcript for

Mar '161

Coding Projects and Duga Heaven :)

Minesweeper Flags, Zomis' games, PDB, MTG, Cardshifter, and al...

browser-tabs duga factorio fleet-command kotlin minesweeper pdb pineapples vue.js

join this room
about this room

00:00

06:00

12:00

18:00

all times are UTC

site design / logo © 2024 Stack Exchange Inc; legal

mobile