/dpt/ - Daily Programming Thread

Previous Thread: >you will never code with cameron
edition

What are you working on, Sup Forums?

Other urls found in this thread:

linuxhowtos.org/C_C /socket.htm
kaggle.com/join/coursera_ml_dota2_contest
interactivepython.org/courselib/static/thinkcspy/index.html
youtube.com/playlist?list=PLGLfVvz_LVvTn3cK5e6LjhgGiSeVlIRwt
automatetheboringstuff.com/
twitter.com/NSFWRedditImage

I hope no one sees this. I don't understand what lambda functions are.

think of them as function literals

Lambdas are for hipsters. OOP is for true developers.

Has anyone masturbated with one of these big floppies?

How do I get a job?

I have a few small scale projects and more are on the way but how do I leverage that into getting employed?

Lambdas and OOP aren't mutually exclusive, especially since OOP languages typically have functions anyway

put your projects on github and show it to people, if you're good you'll get a job, if you aren't you'll whine about it on Sup Forums

what is a project i could do that would impress people, but I could reasonably do in like 20-30 hours?

or you could just answer his question

mylist = []
for x in range(0, 1024):
y = random.randint(0, 1023)

idk make a clipboard manager for windows, that's like 2 hours, every single clipboard manager for windows is absolute shit, i wanted to shoot myself when I was using windows

it was answered you fucking stupid pajeet retard shit
go kill yourself you waste of space

Any interesting ideas for stuff user?

I have a quote getter and a wikipedia searcher and a local weather checker. Nothing too serious. I'm making a calculator and a simon says game as well.

did you or didn't you already know all of the elements in that piece of god? Did you or didn't you? You fucking knew syntax, you knew what it meant and you still couldn't put it together to make the computer do what you want it to do. That's what programming is, and you aren't good at it.

I got a job soon after finishing a chip-8 emulator, writing snake and sokoban in chip-8 assembly. Employers eat up anything low-level and foreign to them

You still didn't answer his question cause he didn't want a sample of x integers in y range

do you know JavaScript? You could make a chrome extension that works so when you bookmark a page it looks what bookmark folder that page is found the most and suggest that folder as the first suggested folder in which you will bookmark your new bookmark. The second folder should be the second most common place bookmarks from that particular page go and so on.

Wanted to make this but I can't bothered with JavaScript. It would save so much time though.

I've only used javascript recently. I don't know much but I can try.

...

Never reply to me again unless you're contributing to the thread.

thanks for the advice user

this is why I don't believe any of you on /dpt/

You were just posting so many questions asking about a chip 8 emulator a few days ago there is no way you got a programming job a few days after that

I was browsing the Cuda documentation when I came across this page.

...

what are you implying?

I got this job in mid-March tho

no problem

Those are the good ones.

you just said you got it after finishing the chip 8 emulator which you asked about a few days ago and now you say you got it in mid March

I WANT TO BELIEVE

I did, I gave a helpful and concise answer that was free of sneering and criticisms

You'll have to tell me how insulting someone is more productive

>don't reply to me again unless

Don't act like a child, this board has age limits

so many people were posting about chip-8 since jan/feb. I just made this general after not posting on /dpt/ in months, so I have no idea what's been going on since

How do I get into programming?Been intrested for a while

ok kid

>read few books
>watch few tutorial courses
>code

there you go

Type text and hope it compiles

Find a guide on BASIC. It is a fun and easy, if archaic, language perfect for cutting your teeth without getting in too deep!

How long would it take me to learn enough programming to get some shitty codemonkey job making $20-30k? I don't care if I have to eat ramen, I just want a foot in the door
>and not to be a NEET anymore
;_;

6 months if you ain't retarded

5 months for me (4 months python, 1 month applying)

can anyone explain this
bzero((char *) &serv_addr, sizeof(serv_addr)); or the (char *) specifically

cast &serv_addr to a char*

suppose I can do 10+ hours a day (I have mild autism and am really good at staying on task)

still a 6-month proposition?
>tfw I can also pretend not to have autism in social settings

also what sorts of projects did you complete while learning/working with python? And does your actual job use python?

Right now I know about basic data structures, objects, APIs, lil bit of python, some objective-c, some swift, I can kinda hack together shitty programs using existing libraries and make them do things, I've been running Linux for like 3 years so I've got some terminal/bash/whatever experience, I know a little bit of Big O/algorithmic complexity stuff

I can write some calculator.app type shit basically

how many months of your estimate does this cut off if any?

thanks guys

it holds the address of serv_addr, and tells the compiler it is an address holding a char type

ancient code, use memset instead.
(char *) is a cast.

2 months of studying and you'll be good

>suppose I can do 10+ hours a day
I didn't do as much, though I updated my github almost everyday during that time

I also finished a brainfuck/ook interpreter, virtual machine implementation of the specs posted on challenge.synacor.com, hold top 10% on codeeval (very easy to do since few people have done more than a few challenges), finished a web scraper for my old job (as a side hobby, as it wasn't programming) and for Sup Forums (obviously, I didn't tell them)

Yes my job is in Python, and also Django, though I came in with no Django experience. Just with my portfolio, I answered 'I don't know' to half the interview questions and still got in

6 months is doable if you're focused

tbf I also have been using C the past 10 or so years, that may have given me a slight advantage, though we don't use C here

how much a year?

I see thanks, I didn't make the code, just reading from linuxhowtos.org/C_C /socket.htm

>2 months
>6 months
oh god

can I average these and assume that four months full time is a safe bet?

>I didn't do as much, though I updated my github almost everyday during that time

thanks for the tip, I suppose I'll start doing that.

I have anxiety about committing shitty code, should I just do it anyway?

I'm not saying while I'm still in probation lol

I created a neural network that given an input of 10 Dota heroes participating in one single match tries to predict how many last hits each will have at 5 minutes.

I used replay data for training from here - kaggle.com/join/coursera_ml_dota2_contest - about 100000 replays.

My model has an embedding layer that maps hero id (there are about 100 heroes) to a vector of 16 dense values, which produces 160 float values from 10 hero ids at input.

Then I have densely connected layer with 256 nodes and sigmoid activation.

Then I have an output layer with 10 nodes and no activation.

Loss function is mean square error. I use Keras to train the network.

I'm not really sure about how well it does with predictions.

What is amazing about it is t-sne visualization of embeddings.

It created very, very strong groups of heroes in dota - and it did that from just data about how many creeps each of 10 heroes in a match killed at 5 minutes.

I now want to try playing with embeddings sizes and also want to try training it on other inputs (bought items instead of heroes) and other outputs (XP, gold, hero kills).

does anyone have a good tutorial for network programming in c for linux

understandable

>should I just do it anyway?
yes

I'm 6 months cause I don't have a degree. Do you have a degree? It might not even take you a month if so

Is there a curriculum I should follow somewhere? Something that tells me the big points to touch on? I know I need to basically just code more but I don't want to blindly work on worthless projects for 6 months and be no farther ahead of where I am now

- algorithms
- searching, sorting
- big O
- getting familiar with libraries/whatever relevant to the job/specific field
- design patterns
- ???

would really appreciate any suggestions here

Some college / no degree, no advanced math

>blindly work on worthless projects for 6 months
just do shit that's interesting

employers probably couldn't care less about the same old code monkey

Beej's guide to networking.

>- getting familiar with libraries/whatever relevant to the job/specific field
>- design patterns

are you greentexting these because they're wrong? I don't know what I'm doing ;-;

>

alright thanks

...

Design patterns are a meme and learning a library is as simple as reading the manual/tutorial the first time you need it.

SHELLEXECUTEINFO sei = { sizeof(sei) };


Can someone explain to me what's going on with this initialization? The brackets? Initializing to it's own empty size? Wut

how are you not sure how well it does with predictions? you should have taken 5-10% of your data and reserved it for a validation set and only trained with the rest of the data
then you can test it by how well it does on the validation set

I see. Can you suggest anything I should add to my little list then?

May I suggest a book? It talk about all the 3 first things you said. It's called "Introduction to Algorithms" by MIT Press.

I'll definitely check it out, thanks. Can you suggest any additional bullet points / main topics?

Well, I do that, but I have no idea how Keras handles predictions. It outputs some value, but it seems bogus. I couldn't even find what it means in documentation.

When I used Tensorflow, I had to define what accuracy is myself. If hero has 25 last hits and model predicted 24, that's pretty accurate. In keras there's doesn't seem to be a way to define what accuracy is and I couldn't even find how it's calculated.

Here's some output:
97115/97115 [==============================] - 0s - loss: 24.1759 - acc: 0.3793 - val_loss: 25.8611 - val_acc: 0.2900
Epoch 473/500
97115/97115 [==============================] - 0s - loss: 24.1727 - acc: 0.3791 - val_loss: 25.7853 - val_acc: 0.3300
Epoch 474/500
97115/97115 [==============================] - 0s - loss: 24.1629 - acc: 0.3795 - val_loss: 25.8477 - val_acc: 0.3100
Epoch 475/500

acc of 0.3795 means the model guessed about 38% of output values, so that seems pretty bad, but since I have no idea what keras considers to be accuracy, I can't interpret it.

Here is a visualization with tow densely connected hidden layers instead of 1. Doesn't seem much better sadly.

The first value of the SHELLEXECUTEINFO structure/array is its own size and the rest is initalised to 0.

int sei = sizeof(sei);
I don't see the problem

>int
why?

scratch that noise! just read this book Practical Programming An Introduction to Computer Science Using Python 3

I have a data structure with name, information, time to complete. What should I name it? Can't really name it activity since it's a lagdroid app, was thinking task/action/step but that doesn't seem to click.

will do

Does Codeblocks support project-wide refactoring?

TaskDescriptor

opinions on this book?

interactivepython.org/courselib/static/thinkcspy/index.html

also for python

>youtube.com/playlist?list=PLGLfVvz_LVvTn3cK5e6LjhgGiSeVlIRwt

>automatetheboringstuff.com/

I see. Weird.

just torrent Think Python O'Reilly

that interactive shit is all broken you'll see as you go along

If I train the model to predict experience instead of last hits, the visualization is much more predictable.

you want to cross-validate
so instead of taking 5% of your data aside and then testing with that at the end you do 10 experiments with a random 5% taken away each time and then average together your accuracy overall

well loss 24.whatever means the mean square error between what your network predicts and what is the reality is about 24, which is also pretty shitty

an accuracy that low means just looking at the 10 heroes in a match does a bad job predicting so try to correlate some things and bootstrap multiple inputs together

cool stuff tho user! the fact that the heros end up being grouped is interesting and means you're somewhat on the right track : )

>a random 5%
the better way is to do full LOOCV

elder titan, earth spirit, oracle, phoenix are supports though

how hard would it be to program a much better league of legends game

>phoenix
>anything but an awesome pos 3 offlaner

kill yourself

>well loss 24.whatever means the mean square error between what your network predicts and what is the reality is about 24, which is also pretty shitty
Eh. sqrt(24) ~= 5; out of maybe 100 last hits total I'm off by 5. I wouldn't call this loss shitty.

I could define my own accuracy function and calculate it after training the model, but I don't need it anyway - I don't want those predictions, I just want trained hero embeddings which I would use in a bigger model.

I often see them as offlaners which puts them between core and support position. Considering they are sometimes full supports and sometimes offlaners I think they fit pretty nicely in the picture.

for training a neural net with 500 iterations and a dataset of 100000? only if you have a cluster of k1's and a whole lot of spare time

both of you plebs at dota

what are your mmrs I bet 1k scrubs

Read SICP

sry, i was going off your example of 25 last hits +/- 5 is 20% lol
you may want to see what the default parameters are for your training methods to see what kind of limits are in place for both the training and accuracy

my mmr is xK, x > your mmr

K = 0

If lua is so much faster than python thon and ruby, why is it not used instead of them, particularly for business?

Seems like it's as easy to write, and faster, I'm not seeing the down side

The libraries and tools are already written in Python and Ruby. Switching to Lua would be a lot of money in investment and time in development.

Lua is incredibly simple and not as expressive as Python or Ruby.
It also doesn't have as much of a strong eco system.

By the way, here are embeddings generated by the big model - one that tries to predict from a lot of data from 5 minutes of a replay which team is going to win.
There are some meaningful groupings, but generally embeddings seem much less organized than specialized embeddings I generated using small models previously. After i find a set of embeddings that satisfy me I plan to place them into the big model and make them untrainable. Right now I predict that outcome of match with 70% probability on a test set.

I'd like us to continue this discussion with emphasis on neural networks and not dota.

it has a jit

sure post your code of how you set up the neural networks and we can discuss your method of how you are doing it