One of the best aspects of your videos is how you walk through the entire coding process. The ENTIRE coding process. You make mistakes and correct them (instead of just giving up.) You say, "Okay, I forget some of this, so I'm going to have to look at the documentation." You run something successfully, but it doesn't do what you want. And sometimes what it does instead is interesting, and worth revisiting later. Great lectures, great stuff.
@@TheCodingTrain Hey Dan, I have a question for you, I am thinking to make a form which when I submit, I want to open a new page with the form details. I am trying this with HTML and JS. I am not sure if even I can do it without php, but still giving it a go. Let me know if you have anything you have come across something related or whatever you think about it.
I'm a Unity game developer and I just watched this whole video for some reason. Your presentation of the problem is so entertaining that I just got lost in it because every word that came out of your mouth was full of energy and had a purpose. Great job!
Sir Daniel, you are the best teacher on UA-cam. I am 13 years old and because of you i know more about programming and AI than even 20 year olds. I am truly so so grateful towards you. Even though i started with python, whenever i have any problem , i look up one of your videos and totally get the topic. You're on of the best programmers around. One of my life's main goals is to meet you, and i hope you never stop making such brilliant videos. Thankyou !
I am a web developer currently starting AI also and i take much knowledge about neural networks and also AI functionalities .. Nice its really awesome .. Thanks :)
@I Understand! For AI i take start from js as i know its flow and its working so i feel easy to start with js .. and also i think that before going on machine level AI i must grip on its basics .. i.e Neural Networks, Data Training etc .. So i also suggest you take start from language you know well else if you direct start from machine language you need to learn it first .. currently i done flappy bird .. Snake game .. and relevant short Ai base programs
So recently i thought to myself: hey Im interested in AI, let's see if I can make a nn myself. Was a bit disappointed after I put several hours in without any real progress. And now Daniel is uploading this gem. Thank you so much, i will definitely watch through and try it on the weekend myself!
oh.. i thought this chapter was over... nice you are back with Neuroevolution!! Can you do something about (deep)Q-Learning and A3C? (more interested in Q-Learning)
@@TheCodingTrain Daniel.. I want to thank you for your hard work. I'm suffering the "Coder's Block"... I wasn't able to understand the most simple algos.. simple codes, anything. And the worst thing, I wasn't able to code.. I got frozen in front of the screen like a plant. This was happening me for about 1 year... So, with a lot of work (reflection and the help of my therapist).. I decided a couple of weeks ago to try something different.. I started from scratch... I found a lot of papers... blogs... code examples... but the most important... youtubers like you that program... And now.. i'm not "cured" 100%... but i can program and understand code again!! Even more complex code that i was able to understand before i got the block. I hope you read this.. You do more than teaching... So thank you Daniel Shiffman :)
the content is great, but I am on the fence on if it is good for learning or not. I start thinking I understand what is going on, then he realizes he made a mistake and quickly changes things, and I am not sure if I am remembering the corrected part or the original. Still great to hear him work through the problems and logic, and while explaining so well.
Thanks for all the videos and tutorials Dan! I went through all of the Nature of Code book, and was really interested in the Neuroevolution section. I made my own version of the flappy bird game a while ago and recently decided to see what a 3d version would look like and how would the neural network behave. Here is what I came up with editor.p5js.org/djuliette/sketches/aGEe50IFd
Great Video!! Suggestions: - Generally JavaScript used camelCase variables and functions, so the variables you created "input_nodes" and "hidden_nodes" should be "inputNodes" and "hiddenNodes". - You could make use of "async-await" feature of javascript to deal with promises and callbacks... e.g. const outputs = await ys.data(); - Why can't you use Object.create(this.model) to create copy new model?
It's so much more complicated than the previous version :D You should also try neataptci.js there's some cool stuff purposely designed for neuroevolution
Finnaly I've been waiting forever for this. I know how it works but I don't know how its made. I'm just really fascinated about how they work with all of the possibilities . I honestly don't know how @the coding train does it but I tip my hat to you. So keep doing what your doing, and I can't wait for the next one.☺
Same here. I know the basics of how neural networks and training works and can implement a crude prototype by myself but I don't know how to use all these librarys.
There wasn't any lesson about weights and bias in your previous lessons. :) Only high-level explanations of them. Would be nice to understand the magic behind all these calculations and what is going on on the background.
do we use reinforcement learning in this video at all? or is it just a genetic algorithm with NN..sorry if sound like an idiot, dont know about this stuff. Thanks !
this is so useful for me. Thank you so much! Could you please go through the slow running speed of tf.js please? I face the same issue too and have no idea how to fix it...
hi daniel, just found your channel recently and I'm loving it. I see you have a great influence on your audience. I recommend you to try out other harder programming languages like python or c++ so people could understand it using your knowledge,if you understand what I'm saying. p.s. you're the best fucking teacher in the entire world I don't think you realise that
I very stuck now with my coursework about genetic algorithms using evolutionary like Using Real-Valued Genetic Algorithms to Evolve Rule Sets for classification, please any guys can suggest the way I should do thank you very much
hello i recently start seeing your videos and are amazing. Sorry about my poor english i'm still learning. If there any chance can you do a procedural generate algorithm like dwarf fortress. Not a complete at that level, but at least a 10%. thankyou
Hey Dan, I have a question for you, I am thinking to make a form which when I submit, I want to open a new page with the form details. I am trying this with HTML and JS. I am not sure if even I can do it without php, but still giving it a go. Let me know if you have anything you have come across something related or whatever you think about it.
same as opening a popup, [3] or clearing the current window[5][6] or cover the current view with an opaque layer[4]. Look to this documentation [1][2] and lastly here's a tutorial to do the evil [3] [1] developer.mozilla.org/en-US/docs/Web/API/Document_Object_Model/Introduction [2] developer.mozilla.org/en-US/docs/Web/API/Window/open [3] html.com/javascript/popup-windows/ [4] www.w3.org/TR/css3-values/#viewport-relative-lengths [5] developer.mozilla.org/en-US/docs/Web/API/Document/all [6] document.open()
love this video. very useful for me. i want to get weight and bias data. is this code correctly for get model data? let weights = birds[0].brain.model.getWeights()[0].dataSync(); let hidenLayerBias = birds[0].brain.model.getWeights()[1].dataSync(); let hidenLayerWeight = birds[0].brain.model.getWeights()[2].dataSync(); let output = birds[0].brain.model.getWeights()[3].dataSync();
Wait, if you're not using cross-over, then why are you not using Simulated Annealing, instead of a Genetic Algorithm? There's really no upside, and SA is both easier to code and has less parameters to tweak in order to get a fast convergence to an optimal solution. SA very roughly looks like a GA with a population of 1, but it's subtly different. Wikipedia has a good article outlining the basic strategy.
I was thinking... "22:36 oh... ok.. it was the console.log......" "31:31 nope... everything related with Python has to be bad.." (I hate Python and its libraries) Thanks for the video, now i know that TensorFlow is bad even in JS. It wasn't enough with Python and C#...
One of the best aspects of your videos is how you walk through the entire coding process. The ENTIRE coding process. You make mistakes and correct them (instead of just giving up.) You say, "Okay, I forget some of this, so I'm going to have to look at the documentation." You run something successfully, but it doesn't do what you want. And sometimes what it does instead is interesting, and worth revisiting later.
Great lectures, great stuff.
Thank you for this feedback!
@@TheCodingTrain Hey Dan, I have a question for you, I am thinking to make a form which when I submit, I want to open a new page with the form details. I am trying this with HTML and JS. I am not sure if even I can do it without php, but still giving it a go. Let me know if you have anything you have come across something related or whatever you think about it.
@@namanmehta5243 See your other comment on this video.
I'm a Unity game developer and I just watched this whole video for some reason. Your presentation of the problem is so entertaining that I just got lost in it because every word that came out of your mouth was full of energy and had a purpose. Great job!
Thank you!
Siraj, this is real machine learning.
Sir Daniel, you are the best teacher on UA-cam. I am 13 years old and because of you i know more about programming and AI than even 20 year olds. I am truly so so grateful towards you. Even though i started with python, whenever i have any problem , i look up one of your videos and totally get the topic. You're on of the best programmers around. One of my life's main goals is to meet you, and i hope you never stop making such brilliant videos. Thankyou !
Watching this in 2020, there is no other channel with the quality of content this channel offers
this is exactly what I was waiting for. Now I know how to use TensorFlow.js layers in a more customised way. Thanks Daniel
I just love the way your lectures flow. So much energy flowing out of the screen.
I am a web developer currently starting AI also and i take much knowledge about neural networks and also AI functionalities .. Nice its really awesome .. Thanks :)
@I Understand! For AI i take start from js as i know its flow and its working so i feel easy to start with js .. and also i think that before going on machine level AI i must grip on its basics .. i.e Neural Networks, Data Training etc .. So i also suggest you take start from language you know well else if you direct start from machine language you need to learn it first .. currently i done flappy bird .. Snake game .. and relevant short Ai base programs
@I Understand! need any help free to talk :)
@I Understand! Node Js
So recently i thought to myself: hey Im interested in AI, let's see if I can make a nn myself. Was a bit disappointed after I put several hours in without any real progress. And now Daniel is uploading this gem. Thank you so much, i will definitely watch through and try it on the weekend myself!
Xy you can do anything you set your mind to
oh.. i thought this chapter was over... nice you are back with Neuroevolution!!
Can you do something about (deep)Q-Learning and A3C? (more interested in Q-Learning)
Eventually I hope!
@@TheCodingTrain Daniel.. I want to thank you for your hard work.
I'm suffering the "Coder's Block"... I wasn't able to understand the most simple algos.. simple codes, anything. And the worst thing, I wasn't able to code.. I got frozen in front of the screen like a plant.
This was happening me for about 1 year...
So, with a lot of work (reflection and the help of my therapist).. I decided a couple of weeks ago to try something different.. I started from scratch... I found a lot of papers... blogs... code examples... but the most important... youtubers like you that program...
And now.. i'm not "cured" 100%... but i can program and understand code again!! Even more complex code that i was able to understand before i got the block.
I hope you read this.. You do more than teaching...
So thank you Daniel Shiffman :)
@@FerMJy I am so glad to hear this!
the content is great, but I am on the fence on if it is good for learning or not. I start thinking I understand what is going on, then he realizes he made a mistake and quickly changes things, and I am not sure if I am remembering the corrected part or the original. Still great to hear him work through the problems and logic, and while explaining so well.
Thanks for all the videos and tutorials Dan! I went through all of the Nature of Code book, and was really interested in the Neuroevolution section. I made my own version of the flappy bird game a while ago and recently decided to see what a 3d version would look like and how would the neural network behave. Here is what I came up with editor.p5js.org/djuliette/sketches/aGEe50IFd
please don't stop with these videos, I love them all, igf you have a son make him tech for you when you get older
Great Video!!
Suggestions:
- Generally JavaScript used camelCase variables and functions, so the variables you created "input_nodes" and "hidden_nodes" should be "inputNodes" and "hiddenNodes".
- You could make use of "async-await" feature of javascript to deal with promises and callbacks... e.g. const outputs = await ys.data();
- Why can't you use Object.create(this.model) to create copy new model?
Thanks for this feedback! I didn't try Object.create(), the question is whether it would be a "deep" or "shallow" copy I think?
If you are not using crossover for evolving in each epoque, then you may be more close to a Immune System algorithm (look for Clonal Selection)
Always a good time when I get the notification for a new vid of yours!
BEST TEACHER EVER, keep going
You are a huge inspiration for me!!!!
Really nice courses thank you for your investment. I will now copy paste your beautiful code in my app and conquer the world.
This video is greatly explained and makes neural networks look easy!
Exactly! We can solve it with if statements. Need a more complex example! :)
It's so much more complicated than the previous version :D
You should also try neataptci.js there's some cool stuff purposely designed for neuroevolution
Finnaly I've been waiting forever for this. I know how it works but I don't know how its made. I'm just really fascinated about how they work with all of the possibilities . I honestly don't know how @the coding train does it but I tip my hat to you. So keep doing what your doing, and I can't wait for the next one.☺
Same here. I know the basics of how neural networks and training works and can implement a crude prototype by myself but I don't know how to use all these librarys.
Exactly
Human moment!
@ 3:43
(thought) "UHHH..." (thought)
Keep doing what you do Daniel! :D
Great one! 😍
A great start to integrate tfjs with neuroEvolotion ,Thank you for your perfect video. can you make another video for implementing "NEAT" in tfjs plz.
Yet another awesome video! Thank you mr. Coding Train. I was wondering thought , is this approach superior to the NEAT algorithm ?
I would say NEAT is probably superior to this? My examples are a simplified, less sophisticated version of NEAT.
This is what I listen to when I go to sleep because coding is seen as relaxing to me
There wasn't any lesson about weights and bias in your previous lessons. :) Only high-level explanations of them. Would be nice to understand the magic behind all these calculations and what is going on on the background.
Maybe u could another series on the NEAT algorithm? I think it would be a good continuation from Neuroevolution
Amazing as always !
Please someone tell me, is it really necessary to clone the weights while the tensors are not mutable ?
Could you do a tutorial on how to save and load TensorFlow.js models, thanks
I really enjoyed this video! Waiting for more
random(1) < rate was the most coolest things :D
do we use reinforcement learning in this video at all? or is it just a genetic algorithm with NN..sorry if sound like an idiot, dont know about this stuff. Thanks !
have you figured this out yet?
Great Sir
this is so useful for me. Thank you so much! Could you please go through the slow running speed of tf.js please? I face the same issue too and have no idea how to fix it...
hi daniel, just found your channel recently and I'm loving it. I see you have a great influence on your audience. I recommend you to try out other harder programming languages like python or c++ so people could understand it using your knowledge,if you understand what I'm saying.
p.s. you're the best fucking teacher in the entire world I don't think you realise that
I believe he's a professor at a university
sanchit verma same
So mutate is just add random 10% number to those weights? Or gradient descent then save those weights after?
I have always enjoyed your videos. Could you please make a video on NEAT!!!!
Could you also do a tutorial on how to show the best performing bird of the current generation, thanks
Beautiful!
Programming ❤️
You are the best
great work
very close to 800k
w + randomGaussian() should it be normalized after that?
Wooooooooooooow cool project! And that too like an hour ago
use the Model class in tf.layers i'ts cool
Do you know how to run Tensorflow in python for modeling and run the flappy bird game in javascript?
pozdrowienia z wiezienia
Hiya your tutorial index doesn't have chapters 11,13,17 as of writing.
considering trying to do this with phaser
Please do! and share!
Awesome video, I love these
Great video!
Wait, you're teaching a class on this? Is that available somewhere?
I very stuck now with my coursework about genetic algorithms using evolutionary like Using Real-Valued Genetic Algorithms to Evolve Rule Sets for classification, please any guys can suggest the way I should do thank you very much
hello i recently start seeing your videos and are amazing. Sorry about my poor english i'm still learning. If there any chance can you do a procedural generate algorithm like dwarf fortress. Not a complete at that level, but at least a 10%. thankyou
I LOVE PYTHON
u crazy smart fam
Hey Dan, I have a question for you, I am thinking to make a form which when I submit, I want to open a new page with the form details. I am trying this with HTML and JS. I am not sure if even I can do it without php, but still giving it a go. Let me know if you have anything you have come across something related or whatever you think about it.
same as opening a popup, [3] or clearing the current window[5][6] or cover the current view with an opaque layer[4]. Look to this documentation [1][2] and lastly here's a tutorial to do the evil [3]
[1] developer.mozilla.org/en-US/docs/Web/API/Document_Object_Model/Introduction
[2] developer.mozilla.org/en-US/docs/Web/API/Window/open
[3] html.com/javascript/popup-windows/
[4] www.w3.org/TR/css3-values/#viewport-relative-lengths
[5] developer.mozilla.org/en-US/docs/Web/API/Document/all
[6] document.open()
You are always great
just curious how would you save and load the weights once I have found the best bird?
I have an example that does that here!
github.com/shiffman/Tensorflow-JS-Examples/ (Look at 04)
@@TheCodingTrain umm i tried ur example but it keeps giving me the error imgur.com/X80aZuw can u explain for me xd
❤
My challenge for you is to create a pong in js file on an html file without the p5.js library
Is there a tutorial on how to make the actual NN?
Maybe try this playlist? ua-cam.com/video/XJ7HLz9VYz0/v-deo.html
is it possible to implement q-learning in javascript with flappy bird environment?
Yes
What company are you working?
You can just mutate your values without making a copy, as you are assigning it back in the end anyway
You do need a copy because of the selection process.
love this video.
very useful for me.
i want to get weight and bias data.
is this code correctly for get model data?
let weights = birds[0].brain.model.getWeights()[0].dataSync();
let hidenLayerBias = birds[0].brain.model.getWeights()[1].dataSync();
let hidenLayerWeight = birds[0].brain.model.getWeights()[2].dataSync();
let output = birds[0].brain.model.getWeights()[3].dataSync();
Wait, if you're not using cross-over, then why are you not using Simulated Annealing, instead of a Genetic Algorithm? There's really no upside, and SA is both easier to code and has less parameters to tweak in order to get a fast convergence to an optimal solution. SA very roughly looks like a GA with a population of 1, but it's subtly different. Wikipedia has a good article outlining the basic strategy.
Thanks for pointing this out! I think one difference here is that I maintain the idea of "generations" and a population of N elements?
Plzz make chain reaction game with p5.js
When's mnist with tf.js
I was thinking... "22:36 oh... ok.. it was the console.log......" "31:31 nope... everything related with Python has to be bad.." (I hate Python and its libraries)
Thanks for the video, now i know that TensorFlow is bad even in JS. It wasn't enough with Python and C#...
\o/\o/\o/ All praise tf.tidy() \o/\o/\o/
Make one that solves a rubix cube m8
it hurts that you realize you are repeating yourself, saying "once again", and not extract the same part into a function