Neural Net implementation in C++

[@rectorsquid]

I was glad to find a tutorial that didn't show "How to use a library to make a neural network so we can skip important details" I wanted to see the low-level construction to help me fill in some missing info in my knowledge and this was perfect.

[@hynesie11]

That's why I searched for C++ neural network. You don't learn anything from using those high level Python libraries.

[@grantallan7355]

This guy has a great "I think you don't know anything so I'll talk down to your level" voice. It doesn't sound like a compliment, but I do mean it as one.

[@Toy_Bubble_7]

I jumped into to something far too complicated in c++ without knowing much of the language, but after following this whole video, when I came back the next day I learned how to use pointers and how classes work. I made test files to get me more familiar with some of the concepts you were using. I am glad I jumped head first into it as I know way more that I would have known about c++ now. The next day I was able to fix issues I created in the program by some of the things I learned. Thank you for making this great video!

[@gautamarora9573]

I am unable to run the last line ./neural-net-tutorial.cpp > out.txt can anybody please help with that, it is failing to run with exit code 1 edit: if anyone in future gets an error try debugging the code first and a possible explanation could be that in my case at least, the txt file made in notepad trainingData.txt stored data as UTF 16 LE instead of UTF 8 so try converting the encoding to UTF 8

[@noiseflow4306]

This might be one of the best tutorials out there

[@cobusuys4787]

This is one of the best hours I have ever spent.

[@ArielLorusso]

0:00 Scope & Requirements 0:51 Neural net Uses and Example 6:00 Net :: ( Class ) 6:47 topology 10:43 inputVals ( Vector of inputs ) 15:52 m_Layers ( Vector of Layers ) 17:26 Net::Net ( constructor ) 23:39 Neuron :: ( Class ) 28:31 Neuron :: Neuron ( constructor ) 30:48 Net :: feed Forward 38:18 Neuron :: set Output Value 38:50 Neuron :: get Output Value 39:11 Neuron :: feed Forward 39:11 Neuron :: transfer Function 44:43 Net :: back Prop 50:57 Neuron :: calc Output Gradients 51:55 Neuron :: calc Midden Gradients 53:00 Neuron :: sumDOW ( Weights Differentials ) 54:05 Neuron :: update Input Weights 55:20 eta & alpha ( Neuron static attributes to adjust learning rate) 58:30 Neuron :: get Results 58:55 TrainingData explained ? 59:18 Prepare for the Superspeed ! 59:26 TrainingData :: object instantiation 59:32 showVectorVals ( function ) 59:34 getRecentAverageError ( function ) 59:34 TrainingData :: (Class) 59:37 TrainingData :: get Topology 59:38 TrainingData :: get Next Inputs 59:38 TrainingData :: get Target Outputs, 59:40 End of Superspeed ! 59:50 set Bias output to 1 1:00:25 Training data file 1:00:32 makeTrainingSamples.cpp 1:00:47 TrainingData.txt 1:01:23 testing the neural-net .cpp ! 1:01:52 output of neural-net.cpp

[@DaJodad]

Thank you very much for this

[@ArielLorusso]

Accounts made 1 & 2 week ago no subs Talking of a non related to this video movie web 80% sure you are bots making publicity on youtube coments ... disproof me

[@dreamhollow]

This was pretty intense, but I now have a far better understanding of neural networks. Thank you.

[@richardrisner921]

Wow!! I have been looking for a walkthrough exactly like this for many months. This got me past the mathematics so that I could get a real working example. Thank you so much!

[@dbr_199]

This is a very high-quality tutorial, thank you so much. Would've loved to see it in higher resolution!

[@techmedia1360]

This is probably the best video I've watched in my life

[@jamesfield5415]

Innit tho

[@FPChris]

Agreed. Amazing video

[@per-axelskogsberg3861]

Yupp

[@coding_with_thomas]

I've been looking around for neuronal network stuff a long time. Found a lot of articles videos and u could save my time by just watching this video. Thank you for this one!!!

[@laureven]

Shame You are busy with work (good thing) :). You are absolutely brilliant as a teacher . Thank you for the video

[@davidnoble5347]

My guy, you really smashed it with this video. Explaining such a complex algorithm, while still managing to not assume your viewers know every nook and cranny of C++ syntax. Well done.

[@ykesav7743]

Great implementation of neural network in C++ STL .Thanks for the tutorial.

[@FatLingon]

"We want to compile often".... goes on to only compile one time during the whole write up.

[@jdm8963]

Don't be greeddy dude. It's obviusly better for us a smooth explanation without sttoping at each modification to compile and test a code that he already said he tested before doing the video. You can pause the video if you want to compile and test.

[@JrPlays]

Yes because the perfect tutorial is expanded with 20-30 more minutes of just compiling after every module is added.... compile it yourself

[@worldshaper1723]

You have changed the world!!

[@majedali1478]

flexing your muscles😄😄😄 by excessive use of memory references Thank you dude really helpful

[@sticktogether2326]

thank you great man =) finally there was someone who wrote a neural network in an adequate programming language p.s. damn python

[@awesomesauce1157]

very true

[@puppergump4117]

Python is not inadequate, it's just barely adequate

[@Uvuv6969]

Yes burn python with fire

[@mustafatuncer4780]

Wonderful course. Thanks a lot.

[@rajatmond]

Invaluable. I can't thank you enough. I had been using TMVA included in CERN's ROOT. But it doesn't really compile in windows. I have the option of openCV but that's bloated. You're a lifesaver

[@FPChris]

Outstanding. Just what I was looking for.

[@kohltonpeterson3287]

He kind of sounds like a programming bob ross and it's amazing.

[@Gabriel-V]

Really nice video. But he (Dave Miller) doesn't explain about the recentAverageError and the smoothingFactor? There are many loose ends, and I had to get them right myself.

[@mustafatuncer4780]

I wish you also prepare a C++ encoder-decoder, GPT project videos.

[@marzipug5439]

I can feel my mind expanding.

[@Hugo-go6yq]

Amazing tutorial thank you so much 🙏🙏🙏

[@Darkl0ud_Productions]

Maybe I'm an idiot, but using Visual Studio and trying to run the neural network with the trainingData.text present, I keep getting an abort() called error. It seems that the program does not recognize my trainingData.text file... Any ideas? EDIT: If anyone is having this issue, it is most likely that you are using UTF-16 formatting for the txt file. I noticed on my Linux machine it would work flawlessly, however in Windows it would not. I realized when redirecting the output to traningData.txt in windows powershell, it outputs in UTF-16.

[@Gauthamphongalkar]

thank you very much!

[@MadMax-mw3og]

Hey! Amazing video, If I could aska question... Would I be able to implement this inside of unreal engine? Anything I need to know before attempting this?

[@f_ftactics7928]

I don't see you use the error in the backprop function after you calculated out.

[@agaveboy]

Thank you

[@mehmetfatih6750]

Hello. In the file on the drive folder, Neuron::updateInputWeights function is deficient. I mean the calculation of newDeltaWeight variable is missing.

[@puppergump4117]

Nobody's gonna mention the fact that this guy woke up on 9/11 and thought it was a great day for machine learning, he must be a robot

[@muhammadzaid308]

I have been vehemntly trying to ignore learning python, and I came so close too

[@pythagorasaurusrex9853]

Fantastic! Exactly what I needed!. Thank you so much!

[@valekprometey]

Great. Thanks a lot for the video. But what is an Epoch in this example? If I know right, backpropagation starts after a whole Epoch completed.

[@Madsycode]

Nice vid though!! i will recommend to make a serie out of this it will bring more subscriber!!

[@tamas5002]

I like this video. Thank you very much for this. Is it mandatory that Neuron class knows about the Network class and vica versa? Couldn't we just simple eliminate this depedency? And, could we out random function into a util class? Because it has nothing to do with a neuron.

[@gauravuttarkar9682]

Is there one for convolutional neural nets?

[@puppergump4117]

I had some issues (probably my fault) with the class Neuron declaration at the beginning. I just removed the declaration and moved the Neuron definition up

[@Tio_kappa]

can you re upload it in 1080p??

[@willr0073]

This was amazing!! Can we use this code? How would we have to reference to it? Thank you!!

[@slubblesthuggy]

Is it normal that it can spit out negative values? Like -0.00234 or something

[@rayenghanem6643]

Hello guys i have a question why is he always using a unsigned and not int in some of the trivial places line for(unisned i ... why not simply for(int i =0... ?

[@TheGoldenriff]

Bob Ross for computer scientists!

[@filakabamba9584]

How can I do the following : Please help asap *1.* Make an Artificial Neural Network with dynamic input and binary out .... *2.* Make a Self Organizing Maps with dynamic input and binary out ....

[@shotakhakhishvili8640]

Does anyone have any idea why this doesn't work for me? I followed his path, I understand what he did very well, but it didn't work. Then I copied the working file, and it still doesn't work, neural network just won't progress at all, it is behaving on test 50 000 just as it was on the first 100 tests. Any idea what I should do? Edit: Test 1 000 000 now, and it "works" just the same, nothing has changed at all...

[@galayuda]

Здравствуйте. Курт Гёдель доказал, что конечно может найтись такой факт который нельзя ни доказать ни опровергнуть, однако это только тогда когда вы ограничиваетесь формализмом и стандартной логикой решения задачи. На формализме и стандартной логике построена компьютерная архитектура. Так как Фон Нейман был одним из приверженцев формазима в математической школе Гилберта. Вполне возможно рутинное решение можно и нужно переложить на компьютер до возникновения "нерешаемой задачи", а уже её решение найти ни применяя компьютер, а применяя так называемый человеческий фактор. Под челвоеческим фактором я подразумеваю фантазию, и воображение, а так же силу творца, которая дарована человеку от создателя. Удачным примером я хочу привести открытие комплексных чисел и решение тем самым текущих задач того времени

[@virozz1024]

how the fuk this gradient derivative is calculated

[@Zi7ar21]

I learned C++ in a very different order lol I still dunno how to use classes

[@waxitoto1234]

they are not that hard, but it took me some time to understand the concept, it's just a struct with extra steps LEL

[@destiny_02]

In c++ struct is same as class, only difference is struct members are public by default and class members are private by default. struct jj { int a; }; Is same as :- class jj { public: int a; };

[@chadgregory9037]

@Jacob maybe you've not really done anything yet that truly required them lol fukin datanerds man

[@sqbi4614]

It just doesn't work for me.

[@sqbi4614]

It is very important to set double newDeltaWeight = // Individual input, magnified by the gradient and train rate: eta * neuron.getOutputVal() * m_gradient // Also add momentum = a fraction of the previous delta weight; + alpha * oldDeltaWeight; with (...) + alpha * oldDeltaWeight and not (...) * alpha * oldDeltaWeight

[@richardrisner921]

@@sqbi4614 I did miss that--thanks! I also found that I needed to set m_recentAverageSmoothingFactor to 100.0 like in the source code. I then initialized instance variable m_recentAverageError. But my program still does not work --it gets stuck at about .46 recent average error with tens of thousands of passes.

[@sqbi4614]

@@richardrisner921 well, stay determined and don't forget to set mutation variable to double (and pass it to functions as double, not unsigned)

[@richardrisner921]

I can compile the source code on his website in C++, so I must have missed something when I was following along in C#. I will have to study this further.

[@richardrisner921]

@@sqbi4614 Thanks!! I got a tremendous improvement with this: replace the approximate derivative 1 - x*x with 1 - tanh(x)*tanh(x). The network as designed was aggressively aiming for binary numbers, and it would get stuck on +1 for my training data. This seems to have allowed it to tune in continuous values.

[@tyforman1821]

crazy white guy #crazies #spook

[@davidporterrealestate]

Where is the original source?

[@krayt5875]

This looks like the original source for the video above www.millermattson.com/dave/?cat=13

[@thechoosen4240]

Good job bro, JESUS IS COMING BACK VERY SOON;WATCH AND PREPARE