The NEAT Algorithm is Neat

Code samples: github.com/Sentdex/NEAT-samples
Neat-Python: neat-python.readthedocs.io/en...
Channel membership: / @sentdex
Discord: / discord
Support the content: pythonprogramming.net/support...
Twitter: / sentdex
Instagram: / sentdex
Facebook: / pythonprogramming.net
Twitch: / sentdex

[@ANTIMONcom]

Fun fact: the python-neat implementation is not limited to evolving neural networks. In my master i used it but with my own custom genome and genes. It is a very general implementation, and the genes can in principal anything as long as you are willing to code some custome gene classes

[@revimfadli4666]

You mean CPPNs and such?

[@ANTIMONcom]

@@revimfadli4666 Yes, i indeed. But a CPPN is sort just a normal network sort of. It was just made in a paradigm/time where it was not normal that nodes used different activation functions, and so it got its own name. (Also it had some bilogical inspirations in its creation) You can also use it for "non nerual stuff". I once used it for points with attributes that was x and y coordinates. (Very simple example, but it was all i needed for evolving substrates for HyperNEAT, witch was what i was doing at the time)

[@revimfadli4666]

@@ANTIMONcom I see, thanks Is it optimized like sharpneat?

[@ANTIMONcom]

@@revimfadli4666 sharpNeat is in C# i think, while this one is in python. I think there is a refactoring currently being done to sharpNeat, but i have never used it. The bottleneck is always the fitness evaluation i feel like, so the optimization of the GA is actually not as important as one might first think actually. Clean-Code-quality wise the neat-python could have been better. But you dont need to look into the code usually. You just give it a population and a fitness function and it does its own thing under the hood. I have done some rewrite of the python neat for myself, but never published it an official branch since it is sort of adapted to spesific things i have done. For example neat-python speciation works on shallow reffrences of individuals in the population. Sort of elegant, but also a bit limiting and confusing. I disagree with this and changed it to holding keys. (I also changed population to a dict, i think it originally was a list when i branched off)

[@apophenic_]

I know none of these words but it is fascinating.

[@ravishankar2180]

This video is gonna be viral in this decade.

[@kuyabobss]

In the next 5 years.

[@User_00128]

it's really neat

[@joshualaferriere4530]

I love how you provided 2 examples to run through. Great teaching!

[@python1108]

Just when I was trying to code NEAT from the paper, your video comes out!!

[@chris-graham]

Box is just a wrapper which applies bounds to the input. It stops silly user errors like [996, 997, 991, 991.2] would output [1.0, 1.0, 1.0, 1.0]

[@sentdex]

ty for the infos!

[@1994ramfan]

You do great work Sentdex! I really like your channel, you provide great explanation on a variety of coding topics. Keep up the great work!

[@sentdex]

Thank you very much!

[@ahmadayazamin3313]

Yay, a new video! That's quite NEAT! Also, how do you actually read research papers? By that, I mean how do you go about understanding it and translating it into code? I usually don't understand what goes on until I read a reference implementation, which is not always available.

[@sentdex]

Depends on the paper. Some are easier than others. Most papers are written for brevity and assume you have a lot of bg knowledge, so I will often spend a lot of time looking up concepts referenced in papers.

[@vspecky6681]

Reading research papers is pretty much a skill. I'm someone who reads papers as a hobby (i.e. I've NOT read tons of papers). Here's how I usually go about it :- 1) Read the abstract and get a basic understanding of what the paper is trying to present. 2) Read the introduction and look for important points. 3) Skim over the "Literature Survey"/"Related Work" quickly. These usually aren't very important unless you wanna actively find out what prior research has been done. You can go over it in more detail after you're finished with the paper. 4) Very carefully read the implementation section since this is the "meat" of the paper. Pay attention to every word, make sure you understand the equations presented. And be ready to actively google stuff you don't understand. This is the section you will spend most of your time in. Even more so if you're gonna actively implement the idea presented. 5) Read the experimentation section to get a grasp on how the algorithm/framework/etc performed on various test cases. This is the section you should come back to when you have finished your own implementation of what's presented in the paper. If you can replicate the test results of the authors with your implementation, you're on the right track. 6) You can read the conclusion and future work sections. I personally just skim em over quickly. 7) See if the paper has an Appendix since Appendices usually contain useful information related to the stuff presented in the paper. If you feel like something about the idea being presented was not explained at all, the authors probably put an explanation in the Appendix. A lot of the time, paper authors upload their code to github. You can take a look at their code if you want a reference. Fun Fact: The first paper I ever read and implemented was the NEAT paper! K. Stanley did a fantastic job on the paper, and it was awarded Paper of the Decade too! I would recommend it to anyone wanting to implement their first (popular) research paper! Link: nn.cs.utexas.edu/downloads/papers/stanley.ec02.pdf

[@digdra]

@@vspecky6681 this is a very good explanation! Kudos!

[@tinatipton3291]

Is there a way to do this where we define observation and actions needing to be taken from the game with image recognition like pyautogui and opencv? Right now we define environment from openai gym, but that doesn't exactly work when screengrab isn't a variable exactly. Also are you able to have it so the algorithm figures out the keyboard on it's own? Or do you need to explicitly define each key on the keyboard? If so, how might someone go about doing this or if anyone can't give me a good answer, do you know how I might be able to go and find the answer to this? As always this is an awesome video, I really enjoyed it and found it really useful for my current project, thank you!

[@kylenwilliams2126]

Are we not gonna talk about the 'closedAI' comment at 4:45 lol

[@vijayabhaskarj3095]

That's exactly what I call them too, look at their new post DALL-E they show us hand picked results, no paper yet, publish stuff only for hype, doesn't release pretrained models, makes money by licencing APIs.

[@FNproject]

@@vijayabhaskarj3095 it's not hand picked

[@vijayabhaskarj3095]

@@FNproject it might not be but we don't know until it's open for to test all variations, now they only allow few combinations.

[@FNproject]

@@vijayabhaskarj3095 "The samples shown for each caption in the visuals are obtained by taking the top 32 of 512 after reranking with CLIP, but we do not use any manual cherry-picking, aside from the thumbnails and standalone images that appear outside."

[@reold]

OpenAI

[@samuelsebastiancornejo2859]

Is there anyway to use these for a number of inputs/outputs that varies over time? other than creating an NN from zero?

[@natonion1154]

I remember NEAT from Sethbling's Mar I/O video!

[@janjunker6458]

You earned a sub, anyone who does anything about NEAT or something similar earns it

[@user-gp8fr1nd3w]

Does NEAT-python work with the 3d environments on OpenAI gym, such as the mujoco?

[@SaviorOfSanity]

Would this kind of algorithm work for a card game like hearts or solitaire, or do those games fall into the "too complex for neat" category?

[@ShadArfMohammed]

Hi, Thanks for this video, I have been trying to implement L-NEAT for classification problems, similar to the one applied by NeuroEvolution of Augmenting Topologies with Learning for Data Classification article. Could you kindly give some ideas on how to implement it in python or maybe a short how to video on it ? Thanks in advance,

[@viktorstefanov4848]

Why do I keep on getting this error when running the thisisneat.py file (I wrote everything 1:1 and even copied everything out of frustration yet to no avail): multiprocessing.pool.RemoteTraceback: AssertionError: [0.13254412211816735, 1.0] () invalid? I tried switching to sublime for pycharm - still no luck.

[@cashbyclose]

so could this be used to analyze past stock market tick data on futures like /ES to predict moves in real time for a Live account.

[@step7steveX]

It is great that you educate people on NEAT! You said that you would improve this implementation by building on top of it. Have you thought how? You should look at doing a video on the evolution of CPPNs using NEAT.. a system called HyperNEAT! Super interesting and useful!

[@shmarvdogg69420]

This video is NEAT squared ;)

[@vihashah5975]

after 23:03, my compiler gives me an error. It says, "module neat has no attribute config" I tried pip installing neat-python a lot of times but it still doesn't work :(

[@karlspencer1493]

It is really awesom to make prototypes on Python. Thanks for introducing me to Neat!

[@joegriffith1683]

at 19:50 you're manipulating the default genome which is how each genome starts. You're genomes may start with 0 hidden nodes, but it will evolve more hidden neurons if necessary.

[@greatsaid5271]

only legends make 2G swap partition when they have 251GB ram :D

[@elginbeloy6205]

Could you do a video covering the implementation of Differential Architecture Search (DARTS)? It is another topology discovery algorithm but it uses annealing to make the prediction function differential w.r.t the architecture then uses gradient descent for optimization.

[@sentdex]

I'll look into it and see, thanks for the suggestion!

[@elginbeloy6205]

@@sentdex Awww! I'm excited! :) Also, thank you so much for responding! I love your videos, and my first intro to Python was through your game series when I was a kid. Many thanks! Love the content!

[@herobrineminecraft1579]

What would the reward for the MountainCar-v0 be? Can anybody help with the reward code?

[@kbisht3680]

Love your videos. Are you planning on continuing the neural network from scratch playlist?

[@sentdex]

Yes

[@kbisht3680]

@@sentdex wow thabk you so much. I have been looking forward to it

[@samselvaraj8171]

what library did you use for the conways game of life visualization?

[@sentdex]

Iirc it was just opencv.

[@samselvaraj8171]

@@sentdex thankss!

[@ojussinghal2501]

I wait for that random laughter in every video. It is so NEAT!

[@irem.6784]

you have to do record the optical character detection in pycharm with qt designer interface , please

[@Argoo41]

May I ask you where do you find you clients fo ML and who are they? Hey from Russia. Thanks for content and shared knowledge

[@rpraver1]

How about with a fully connected net works?

[@pieronexus4709]

Almost at a million subs, nice!! Also, others might have asked this, but can you give any provision when the GTA 5 self driving series will continue? If it will continue.

[@epicquad101]

Is the pdf naturally dark? If not how do you get to view pdfs in dark mode? Thanks for the vid!

[@sentdex]

Its probably the browser extension called "dark reader" It essentially dark modes everything online

[@Stinosko]

I wonder 1) can neat algorithm be used for real time learning? For example with the pole environment, every time it run a game it adjusts the network without needing to live in batches and populations? My initial impression of this implementation it can't? 2) can you feed in a starting networking and improve that network. For example for the detection of hand writing: learn the initial complex part by a reinforced or conventional algorithm on a online databases of hand writings. Than use that network for neat with a new (smaller) batch of images of your own handwriting quicker?

[@sentdex]

1) You could, as long as you have a way to allow for parallelization/many agents simultaneously. 2) the NEAT algorithm in general would allow for this, not sure how it'd be done in the neat-python package, but I am also not too sure how well it'd work/play out, or if this case of transfer learning would outperform pre-existing methods of quickly transfering to a new dataset.

[@Stinosko]

1) i was more looking in terms of live data streaming and learning. But i guess that requires different kind of implementations or different algorithms. For example: streaming data from a game and progressively "learn" the game were you could only have one simulation at a time 🙂 2) Cool, would be nice to see how cpu performing the reinforced learned network is vs the adjustmented neat (after all training). Neat has probably less neurons to calculate as it can get rid of unused neurons but probably doesn't have the performance advantages of big arrays and gpu acceleration to speed it up (yet) 🤷‍♂️

[@cwirz]

@@Stinosko I was trying to use neat-python for the Halite bot game and also wanted to implement some kind of learning by doing approach since the bot could not be updated after submition. I could not get this to work howevery but i also think that you have to update the neat-python package internals to get this done. I am very interested in this if you have any results or ideas. :)

[@super0spore0fan]

Maybe try exploring rtNEAT, stands for real time NEAT.

[@deadcrypt]

NEAT is the best, awesome to see you covering it

[@Nimkrox]

When using 'CartPole-v0' and 'CartPole-v1' it worked nicely for me. But once I've changed the environment to 'BipedalWalker-v2' I got an error ❌: "gym.error.DeprecatedEnv: Env BipedalWalker-v2 not found (valid versions include ['BipedalWalker-v3'])" So I've changed the environment to ''BipedalWalker-v3' and another error occurred ❌: "Failed building wheel for box2d-py" I've searched it on the Internet and found out that a solution was to install package 'box2d-py'. 💡 Unfortunately, when doing this I got yet another error ❌: "ERROR: Failed building wheel for box2d-py" So in the end I had to first install 'swig', then 'box2d-py' and only then 'BipedalWalker-v3' worked for me 😁

[@ayush0477]

this video is the final push towards 1 millions subs.

[@Djellowman]

imo, for lots of environments/problems, it's a lot faster writing your own bot rather than setting NEAT with an interface, the config file, changing your game/problem code to accommodate the algo... and the documentation is very bad. I struggled a bit to get it working.

[@bilybak2]

Can you make video about programing NEAT from scrach?

[@sentdex]

Probably won't, but if you want to you can read the literature or break down the neat python package if you're curious enough

[@prabhat_3549]

U r very close to 1M subscribers,,hope you May hit it soon

[@umairnasir4024]

Great Video, nice work! I do want to add that there's a high probability that NEAT's successors can beat AlphaStar and deepRL performances.

[@umairnasir4024]

This was great but the emphasis on deepRL is too much but you can't really compare these algorithms. Maybe quality diversity which is way powerful can actually be compared.

[@hoaxuan7074]

If you add a bunch of numbers you are doing a dot product! The vector dot the bunch of numbers in vector form. Likewise adding and subtracting a bunch of numbers is a dot product. For example dot the bunch of numbers. Using certain patterns of adding and subtracting you get the fast Walsh Hadamard transform. Which you can view as a system of orthogonal dot products. Where the cost per dot product is a very low log2(n) add subtract operations. Anyway for NEAT you can look up the Continuous Gray Code Optimization algorithm pdf online. The mutation function is very simple but effective. Random + or - a.exp(-p.rnd()). If the parameter being mutated is constrained between -1 and 1 then a is 2. p is the so called precision of the mutation and is positive. rnd() returns a uniform random between 0 and 1.

[@mrfrozen97-despicable]

Worked with neat on different games 1) works well for Flappy bird 2) works well for ping pong 3) good enough for racing game (2d car racing) 4) really bad for snake game ...I think it would take tens of thousands of generations to teach neat nn to play snake.... Maybe I am doing something thing wrong. Anyways . Thanks sentdex)))) learned new things ....... Ml is so fun

[@robinferizi9073]

Why are the waits between each NNFS videos so large, I’m dying to find out how to do it all. I have the book but it’s so much easier to understand from the videos. Especially with your explanations and the animations

[@sentdex]

The videos take 2 people to make rather than just me, and we both have other things to do besides just the NNFS videos, and finally the NNFS videos take a lot of work to produce, especially with the animations, but also a lot more care and effort goes into them to make sure they're good.

[@robinferizi9073]

@@sentdex Right yeah that makes sense, I’d also like to thank you for this series, it’s so easy to follow and satisfying to learn to code

[@ryanalvarez2926]

You might like the MAP-Elites algorithm. It builds off NEAT :)

[@user-nn7mg3bp4u]

i love it, this guy snowden

[@jacewalton6677]

He DOES look like Snowden!

[@user-nn7mg3bp4u]

@@jacewalton6677 they should do a podcast together

[@hoaxuan7074]

Training algorithms can only hope to explore the full solution space of small neural nets. With large conventional nets its seems you only get to explore a much smaller space of statistical solutions. No single neuron can be so exceptional. Neurons have to work together in statistical groups. You can see this clearly with Fast Transform fixed-filter-bank neural nets where each neuron is forcebly connected to all the neurons in the previous layer. Then statistical behaviour is the only thing possible.

[@revimfadli4666]

By "all neurons in the previous layer" you mean past timesteps in the filter? Because "all previous layers" sounds like a cascaded network

[@MultiNarutoGamer]

I am not sure if I understood it, but can this algorithm help to find the optimal number of hidden layers in a CNN?

[@sentdex]

No that's not really the use-case here. You might be interested in something like keras-tuner for that: pythonprogramming.net/keras-tuner-optimizing-neural-network-tutorial/

[@MultiNarutoGamer]

@@sentdex Thanks!

[@alikamel6025]

Amazing as usual.

[@mrfrozen97-despicable]

Tears of happiness from my eyes...... Ah thanks.............)))))))

[@SkyFly19853]

The Legend has returned... ✅✅✅✅✅

[@programming_hut]

your way teaching is very great 👍

[@sentdex]

Glad you think so!

[@gumbo64]

I made a thing with this package only about a month or two ago!

[@mrfrozen97-despicable]

1million + views totally deserved!!!!! Help guys

[@mrfrozen97-despicable]

@wise guy neat algorithm. Google it

[@gamingbugs9296]

Same code but stuck on ****** Running generation 0 ****** Any help?

[@sentdex]

Are you doing walker or cartpole? What kind of processor? Might need to lower population. Try doing population of 25 and go from there

[@gamingbugs9296]

@@sentdex I just figured out that when running on jupyter notebook it doesn't work but running on cmd it works fine. Btw am doing cartpole.

[@rpraver1]

So when can we expect NEFS (Neuro Evolution From Scratch)? Fully connected and NEAT... Your last book is great. I can hardly wait....

[@adempc]

Thanks -

[@jbelisario001]

"closed ai doesn't really help anymore"..........HA

[@nabil.hamawi]

that's sad

[@AChadi-ug9pg]

Hello ! Would u do a series of tuts on MARL (Multi agent Reinforcement learning) plz U r the best

[@sentdex]

Possibly, never heard of.

[@AChadi-ug9pg]

@@sentdex kzbin.info/www/bejne/r6HSZYWafaimepI kzbin.info/www/bejne/oaDToH-wq81ooLs

[@UnPetitPoulet]

There I was thinking about a way to automaticaly build a NN and you come in with that bad boi you know what's coming N E A T

[@sentdex]

Neat timing :D

[@adeeb12321]

that's neat!

[@abhijitbhagavati2421]

Can someone help me run NEAT training on my GPU?

[@ayaanp]

This video about the NEAT Algo is really NEAT, especially coming from a NEAT person ok that was a lot of puns xD

[@pythoncure6755]

Pretty Neat

[@ibrahimkhurshid129]

Me waiting for sentdex 999k to 1M.

[@Kinetic_CGI]

Someone needs to run this on a raspberry pi and combine it with blender! But first I'm going to make it play Galaxian (1983) and Centipede (1982).

[@SkyFly19853]

Hopefully, you will make brand new tutorials on Python, 2d and 3d...

[@sentdex]

Dno what those are

[@SkyFly19853]

@@sentdex the ones from 3d OpenGL, etc.

[@SkyFly19853]

@@sentdex I am very interested in 3d for making video games. Python has so much potential for Video Game development.

[@szymoniak75]

Wait do you have 256Gig of RAM and 48 cores?

[@sentdex]

yeah, this is the boxx machine: kzbin.info/www/bejne/bF6slXV9nM-iaa8

[@szymoniak75]

@@sentdex holy smokes, I could store my entire filesystem on that vram

[@harinkumar1073]

cool beard, mr.snowden

[@arjunkashyap8896]

I like my algorithms as I like my alcohol.. NEAT

[@ahmedelnakeeb5853]

plzzz continue neural network from scratch series

[@sentdex]

That series will also continue, it's just not going to be the only thing this channel does for a year :P

[@ifmondayhadaface9490]

20:25, that is really fast typing.

[@sentdex]

I did recently discover that I have a fairly exceptional WPM.

[@ifmondayhadaface9490]

@@sentdex I was all happy because I got 70 wpm but that looks to be 100 wpm atleast. question about the neural networks from scratch series. What kind of math level is required in order to not get lost? I’m assuming a prerequisite of Algebra 1 judging from the last few videos. But I’m just asking for when we get into stuff like backpropagation.

[@randallwalkerdiaz1002]

I got my friggin book bro.. So excited about the future of humanity where anyone can save up 80 bucks and learn how to make artificial intelligence. My future looks super bright!

[@arcturus1140]

15:44 we NEAT to

[@seth8141]

NEATs unite

[@xantos-zv2ux]

This video is neat

[@karlspencer1493]

Nice

[@karlspencer1493]

Or dare I say "NEAT"

[@sahilkirpekar]

*Bri-ish accent feedback be like:* Yes. Yesh. Yes. Very Neat. Very Neat. ❤️❤️

[@gardnmi]

vscode devcontainers is the best way to replicate a coding environment on any platform that I have found so far.

[@burakalpyasar755]

Lol that title

[@pyhlo9201]

400+ title

[@hippzhipos2385]

Good jokes guys, good one

[@DamienLaRocque]

Is this Genetic Algorithms ?

[@ishaandugar7896]

The video was NEAT.

[@skilz8098]

I'm using Python 3.8 running it on Windows 7 x64. I was able to get the CartPole to work fine. When I started to try the environment for the BipedalWalker I'm getting this message from Python: D:\Dev\Languages\Python\Projects\Neural Net>python env-test.py Traceback (most recent call last): File "env-test.py", line 4, in env = gym.make("BipedalWalker-v3") File "C:\Users\skilz99\AppData\Roaming\Python\Python38\site-packages\gym\envs egistration.py", line 145, in make return registry.make(id, **kwargs) File "C:\Users\skilz99\AppData\Roaming\Python\Python38\site-packages\gym\envs egistration.py", line 90, in make env = spec.make(**kwargs) File "C:\Users\skilz99\AppData\Roaming\Python\Python38\site-packages\gym\envs egistration.py", line 59, in make cls = load(self.entry_point) File "C:\Users\skilz99\AppData\Roaming\Python\Python38\site-packages\gym\envs egistration.py", line 19, in load fn = getattr(mod, attr_name) AttributeError: module 'gym.envs.box2d' has no attribute 'BipedalWalker' I tried doing: pip3 install gym[box2d] however, it gives me this message: D:\Dev\Languages\Python\Projects\Neural Net>pip3 install gym[box2d] Defaulting to user installation because normal site-packages is not writeable Requirement already satisfied: gym[box2d] in c:\users\skilz99\appdata oaming\python\python38\site-packages (0.18.0) Requirement already satisfied: scipy in c:\users\skilz99\appdata oaming\python\python38\site-packages (from gym[box2d]) (1.6.0) Requirement already satisfied: numpy>=1.10.4 in c:\users\skilz99\appdata oaming\python\python38\site-packages (from gym [box2d]) (1.19.5) Requirement already satisfied: pyglet=1.4.0 in c:\users\skilz99\appdata oaming\python\python38\site-packages ( from gym[box2d]) (1.5.0) Requirement already satisfied: Pillowgym[box2d]) (0.18.2) Using legacy 'setup.py install' for box2d-py, since package 'wheel' is not installed. Installing collected packages: box2d-py Running setup.py install for box2d-py ... error ERROR: Command errored out with exit status 1: command: 'c:\program files\python38\python.exe' -u -c 'import sys, setuptools, tokenize; sys.argv[0] = '"'"'C:\\Use rs\\skilz99\\AppData\\Local\\Temp\\pip-install-z0m_nl9y\\box2d-py_e38b54f40db74c9eaab0fc3a6aa8e87f\\setup.py'"'"'; __fil e__='"'"'C:\\Users\\skilz99\\AppData\\Local\\Temp\\pip-install-z0m_nl9y\\box2d-py_e38b54f40db74c9eaab0fc3a6aa8e87f\\setu p.py'"'"';f=getattr(tokenize, '"'"'open'"'"', open)(__file__);code=f.read().replace('"'"' '"'"', '"'"' '"'"');f.clos e();exec(compile(code, __file__, '"'"'exec'"'"'))' install --record 'C:\Users\skilz99\AppData\Local\Temp\pip-record-p933 ly3o\install-record.txt' --single-version-externally-managed --user --prefix= --compile --install-headers 'C:\Users\skil z99\AppData\Roaming\Python\Python38\Include\box2d-py' cwd: C:\Users\skilz99\AppData\Local\Temp\pip-install-z0m_nl9y\box2d-py_e38b54f40db74c9eaab0fc3a6aa8e87f\ Complete output (16 lines): Using setuptools (version 49.2.1). running install running build running build_py creating build creating build\lib.win-amd64-3.8 creating build\lib.win-amd64-3.8\Box2D copying library\Box2D\Box2D.py -> build\lib.win-amd64-3.8\Box2D copying library\Box2D\__init__.py -> build\lib.win-amd64-3.8\Box2D creating build\lib.win-amd64-3.8\Box2D\b2 copying library\Box2D\b2\__init__.py -> build\lib.win-amd64-3.8\Box2D\b2 running build_ext building 'Box2D._Box2D' extension swigging Box2D\Box2D.i to Box2D\Box2D_wrap.cpp swig.exe -python -c++ -IBox2D -small -O -includeall -ignoremissing -w201 -globals b2Globals -outdir library\Box2D -k eyword -w511 -D_SWIG_KWARGS -o Box2D\Box2D_wrap.cpp Box2D\Box2D.i error: command 'swig.exe' failed: No such file or directory ---------------------------------------- ERROR: Command errored out with exit status 1: 'c:\program files\python38\python.exe' -u -c 'import sys, setuptools, tok enize; sys.argv[0] = '"'"'C:\\Users\\skilz99\\AppData\\Local\\Temp\\pip-install-z0m_nl9y\\box2d-py_e38b54f40db74c9eaab0f c3a6aa8e87f\\setup.py'"'"'; __file__='"'"'C:\\Users\\skilz99\\AppData\\Local\\Temp\\pip-install-z0m_nl9y\\box2d-py_e38b5 4f40db74c9eaab0fc3a6aa8e87f\\setup.py'"'"';f=getattr(tokenize, '"'"'open'"'"', open)(__file__);code=f.read().replace('"' "' '"'"', '"'"' '"'"');f.close();exec(compile(code, __file__, '"'"'exec'"'"'))' install --record 'C:\Users\skilz99\A ppData\Local\Temp\pip-record-p933ly3o\install-record.txt' --single-version-externally-managed --user --prefix= --compile --install-headers 'C:\Users\skilz99\AppData\Roaming\Python\Python38\Include\box2d-py' Check the logs for full command o utput. I'm not sure how to resolve this any tips, helps or suggestions would be much appreciated!

[@prabhat_3549]

This video is so neat Corona virus :(

[@SimonZimmermann82]

Anyone seen NEAT with LSTM?

[@ANTIMONcom]

10:18 "there needs to be some kind of API", yes and no. There is a lot of code to copy paste to just run it, but that is because the python-neat is not really first and formost a Neural Networks package, it is a genetic algorithm package (dont know if that was the intent, but thats how they coded it). Thus it expects you to define the genes and how they should behave. The pacgage then runs this in its evolutionary system. But nodes for nurons and weights follows with the package, and i have never seen anyone else use custome genes with python-neat, so maybe an easier API could be build on top of it just for Neural Networks

[@Stinosko]

NEAT video 🤠

[@shmarvdogg69420]

You beat me to it.

[@Stinosko]

@@shmarvdogg69420 i liked your comment in return 😉

[@rookie2961]

Omg! 😍👏

[@ANTIMONcom]

The type of problem NEAT is often used for is reinforcement learning, so comparing it to supervised learning like gradient decent/ deep learnings a bit strange. (Yes i know they both operate on Neural Networks, but one takes a but ton of data with labels, and the other a but ton of tries on the problem.) They have different strength, weaknesses and requirements. The places you might use NEAT, supervised learning might not be an alternative. Maybe this is what you were trying to say and i just missunderstood 🤔

[@revimfadli4666]

PPO, DQN, SAC and other "mathematical" RL algos also use backprop/gradient descent and can be applied to deep networks. Gradient descent doesn't always mean supervised learning. Even GANs and autoencoders use backprop

[@pervezbhan1708]

i think you can work for google, Microsoft and other high tech , if not what is problem and how for other who follow you really understand your logic for understanding the machine learning.

[@sentdex]

No shade to anyone who is, but I am not interested in working for any of those companies at this time.

[@pervezbhan1708]

@@sentdex your answer is same as deep learning

[@brianrichardson1117]

Aren't you trying to solve a simpler problem than was originally intended? Your software merely keeps the pole upright from an upright position, whereas most other attempts are solving the problem starting with the pole hanging. The solution involves moving the slider back and forwards just to get the pole vertical and then balance it. Much harder and may explain some of the "redundant" code. Its basically a simpler version of the car and two hills problem.

[@yousefnashwan3366]

Are all of the comments puns to the word neat? That's not NEAT.

[@shaleenbansal5107]

You look like Jese Pinkman

[@joeljohn2679]

Why can't we reach 1 MILLION subs!! That would look NEAT

[@sentdex]

soon!

[@joeljohn2679]

@@sentdex I mean with the best Machine Learning tutorials on KZbin, you truly deserve more than a Million subs

[@arcturus1140]

Sadly none of the code works for me it's a very good tutorial aside from that

[@c.k7715]

Pls try stock trading with reinforced learning. Ps nice Vids

[@darkseid6412]

i sent you an email, did you see it? edit:k

[@meta_ai]

Boy you made so many mistakes. Please read the paper. I trust you. I like your content.