Who would have thought that a person could relatively understand ML in a span of a week watching some videos. You are an excellent teacher. I now have enough basic understanding to explore. Thank you very much for your time and effort. MY XOR VISUALIZATION WORKED!
Hi! I'm 9th grade student of software engineering in Durango, México and now I have the subject "intelligent systems" (basically "neural networks and machine learning") and thanks to you I'm learning a lot faster than my classmates I've already watched the neural networks series and know I'm learning about tensorflow so idk I just want to thank you for create a great content it helps a lot and I hope this message help you to keep motivated 'cause I'm in México and I always recommend your videos and I know that I'm not the only one in the country so you're global dude =).
I honestly watched until the end and it's totally kinda inspiring how you do all the stuff's. Hopefully I will start with tf js soon, I just need to learn some basics of neural network maybe I will endup watching your series on that. Thanks for the video.
Hi Daniel, the newest tensorflow forbids concurrent calls to model.fit () I wondered when I saw the video, how these many calls to the fit function behave
@@TheAkira2023 I solved it using a semaphore. I.e. just a boolean. Before you call the fit method you set it to true, after the fit to false. All calls to the method, that are invoked, while the semaphore is true are ignored. Switching back to an older version just helps to ignore the problem (which would be there nevertheless, even though older versions don't throw an error).
if you change the "train function" like this: async function train(){ await trainModel().then(result =>console.log(result.history.loss[0])); setTimeout(train,100); } it'd work, and even you can replace timeout function with train(); that will also works.(because it is just waits) I figured out this after watching "promises/await" videos. Thanks Dan.
You said that "a single perceptron cannot solve the xor problem"....which is false, the correct statement is "a single sigmoid perceptron cannot solve the xor problem." A single perceptron can have many different types of "annealing" applied, as sigmoid has been the most common, I can understand the fallacy....but it's not the only type. One that I prefer to use is the normal distribution curve method, where the data has to be at a specific value to give the highest result, and the result goes to zero as you go away from it in both directions....this is much closer to an actual neuron in my opinion as it is easy to see neurons "syncing" with datasets faster for memorization. The formula for it is: simplistic = 1-|B-x|/R curved = e^(-((x-B)/R)^2/2) Where B and R are also trainable sets, while X is determined through the normal sum(An*Wn) method. Xor can be solved with W0 = 1, W1 = 1, B = 1, and R approaches 0.
Hey Daniel I was thinking that maybe you could start a series which explains graph traversal algorithms and their implementation! Keep up the great work!
Also, I am no expert on Machine learning, but there is something called overlearning its where training data isn't used in batches and it over learns the data. I notice you did not mention this!
Is tensorflow better for sequential prediction of 1 array, to classify newly added integers or labels. I am trying to predict random generated numbers. I would like for it to predict up to 8 numbers next in the series. I spent a week learning this and ml5 isn’t working that well. I need more accuracy than LTSM and ChaRNN. I hate python, so I think time series is out of the question.
Can you try Chemotaxis of E.Coli. So use a Neural Network and see if your bacteria can find food just by variing their tumblimg frequency and (show "perfect adaptation"). Or did you do videos explaining how complex a NN has to be to solve a paticular problem? (I know a single perceprton cant solve XOR :) ) I tried Chemotaxis myself with a one-hidden -layer NN, but didnt get good results. So I asked mysel if the NN is to simple for that?
When I took out noLoop(); after I implemented model.fit() it never works. It says it can't start another fit() function even though I am using async/await
You need to ensure that the last fit is done running. Continue the video to see how he moves the rerunning of it to .then(() => {}). My guess is that in the version of tensor flow he is running there is no error output for the error (even though it is there). Google probably added the error later, to ensure fit is not run multiple times at once. :)
Was wondering if anyone had some advice, i am looking at entry level programming jobs and watching these tutorials although fun, is kind of discouraging. This guy is so fluid in his coding and i cant help but feel like crap thinking this is the level i need to be on, and i feel no where close.
You've stumbled on one of the less useful, more complex (and not necessarily in a good way) videos. Maybe try these playlists to start? You can do it! czcams.com/users/shiffmanplaylists?view=50&shelf_id=14&sort=dd czcams.com/users/shiffmanplaylists?view=50&shelf_id=2&sort=dd
Hey in the linear really regression challenge everything is working but stuck at the tf.train.sgd is giving error i have rechecked 100 of time but didn't solve the problem
Daniel, could you make a game where there's no friction and the player never slows down, but it can accelerate in the direction you turn the player in. So for example, you turn the player 90 degrees around it's own center, and when you press the up arrow, the player moves towards that direction. I tried doing this but i'm really stuck, so any help would be appreciated!
Thx a lot for all your videos. I don't look at all your live streams, but I try to look at all your edited videos and I learn a lot of thinks. Thx :) Would it be possible to have a genetic algorithm with tf.js ? If so could you remake your flappy bird genetic AI with tf.js please ? Love & gratitude.
the last 5 were all TensorFlow. and he's more than welcome to enjoy it, but I don't want to see him start to lose subs because he's not changing things up anymore. (not saying *I* would either) but it's something that happens a lot. people get comfortable with something and stop pushing themselves.
He ain't going to lose any subscribers which aren't that important, views are important which keep increasing, also TensorFlow is the new hotness and will be for quite a while, also it's a whole new field for him, of course it's hyper interested. Basically, suck it up man, it's not going anywhere anytime soon.
Thanks for this feedback! I've heard the same from others and took a break from the ML stuff for today's stream. I will be focusing a lot on ML this summer, but it's not my plan for it to be the primary topic for the channel.
Who would have thought that a person could relatively understand ML in a span of a week watching some videos. You are an excellent teacher. I now have enough basic understanding to explore. Thank you very much for your time and effort. MY XOR VISUALIZATION WORKED!
8:19 your tutorials are incredible and you somehow manage to make me laugh in the oddest places
27:30 is like my favourite bit of any coding challenge ever
Wecleome to the Dancing Train!
@@offtheball87 ahahhah I've watched this like 10 times
@@lukegarrigan5894 Thank you for bringing this back up, it'd been far too long!
please watch at 0.25x
@Tommy Kannon don't think I won't report you for spam as well
Absolutely love these tensorflow tutorials/challenges.
I'd love to see more real world examples eventually!
Hi! I'm 9th grade student of software engineering in Durango, México and now I have the subject "intelligent systems" (basically "neural networks and machine learning") and thanks to you I'm learning a lot faster than my classmates I've already watched the neural networks series and know I'm learning about tensorflow so idk I just want to thank you for create a great content it helps a lot and I hope this message help you to keep motivated 'cause I'm in México and I always recommend your videos and I know that I'm not the only one in the country so you're global dude =).
Wonderful, thank you! Check out the Math2Me channel btw, I just met them this week they are amazing!
I honestly watched until the end and it's totally kinda inspiring how you do all the stuff's. Hopefully I will start with tf js soon, I just need to learn some basics of neural network maybe I will endup watching your series on that. Thanks for the video.
I watched it all and loved it. Lol Adam optimizer at the end.
Hi Daniel, the newest tensorflow forbids concurrent calls to model.fit ()
I wondered when I saw the video, how these many calls to the fit function behave
I had the same issue and I don't actually know how to overcome that. But you can always use the older versions as he was using 0.11.2. GL
@@TheAkira2023 I solved it using a semaphore. I.e. just a boolean. Before you call the fit method you set it to true, after the fit to false. All calls to the method, that are invoked, while the semaphore is true are ignored. Switching back to an older version just helps to ignore the problem (which would be there nevertheless, even though older versions don't throw an error).
if you change the "train function" like this:
async function train(){
await trainModel().then(result =>console.log(result.history.loss[0]));
setTimeout(train,100);
}
it'd work, and even you can replace timeout function with train(); that will also works.(because it is just waits) I figured out this after watching "promises/await" videos. Thanks Dan.
@@ForChiddlers thank you so much
We need more of these lessons! Please :)
Loving the TFJS videos... Not many people doing good introductions to it right now.
I like your videos. Thank you for your time.
Thanks! This is all so easy and cool!! :D
I love the way you teach. I'm learning ML through your CZcams channel. Please when can I expect a video on Convolution Neural Nets?
Later this summer I hope!
Great!
@@TheCodingTrain can't wait!
Man.... I just love you, you are amazing
Because of you I switched to p5.js. Hahaha
Watching from playlist session 6. Is this the last of the tensor flow series?
That was really #something.
I was hoping Iris dataset. But oh well. thanks Dan!
Amazing video as always!
You said that "a single perceptron cannot solve the xor problem"....which is false, the correct statement is "a single sigmoid perceptron cannot solve the xor problem." A single perceptron can have many different types of "annealing" applied, as sigmoid has been the most common, I can understand the fallacy....but it's not the only type. One that I prefer to use is the normal distribution curve method, where the data has to be at a specific value to give the highest result, and the result goes to zero as you go away from it in both directions....this is much closer to an actual neuron in my opinion as it is easy to see neurons "syncing" with datasets faster for memorization.
The formula for it is:
simplistic = 1-|B-x|/R
curved = e^(-((x-B)/R)^2/2)
Where B and R are also trainable sets, while X is determined through the normal sum(An*Wn) method.
Xor can be solved with W0 = 1, W1 = 1, B = 1, and R approaches 0.
Thanks for this clarification!
Hey Daniel I was thinking that maybe you could start a series which explains graph traversal algorithms and their implementation! Keep up the great work!
He did! =D czcams.com/play/PLRqwX-V7Uu6bePNiZLnglXUp2LXIjlCdb.html
Indeed!
You are a good friend to anyone
This is what internet should be used for
how about making a trainModel() function which generates a random model of a train and exports it to OBJ
also, #something
hope to see more about tensorflow.js courses
released a whole new set today!
Also, I am no expert on Machine learning, but there is something called overlearning its where training data isn't used in batches and it over learns the data. I notice you did not mention this!
Yes, this is called "overfitting"!
Is tensorflow better for sequential prediction of 1 array, to classify newly added integers or labels. I am trying to predict random generated numbers. I would like for it to predict up to 8 numbers next in the series. I spent a week learning this and ml5 isn’t working that well. I need more accuracy than LTSM and ChaRNN. I hate python, so I think time series is out of the question.
Why is the model in some cases learning the center of the canvas, meaning (0.5, 0.5), be true (like 48:14) and sometimes be false (like 49:52)?
Hi why do we need to shuffle the training data?
That was amazing tutorial, learned a lot. Thanks. #tensorflow
are the tensors disposed with the way you implemented with the setTimeout?
50:52 love how he casually ignored how the model just output all 0's there
Can you try Chemotaxis of E.Coli.
So use a Neural Network and see if your bacteria can find food just by variing their tumblimg frequency and (show "perfect adaptation").
Or did you do videos explaining how complex a NN has to be to solve a paticular problem? (I know a single perceprton cant solve XOR :) )
I tried Chemotaxis myself with a one-hidden -layer NN, but didnt get good results. So I asked mysel if the NN is to simple for that?
Why don't you provide tutorials for Raspberry pi python programming? Btw love your videos
Hey, can you make a video on Pose detection for hand. It would be very helpful. Thanks
Are you on LSD dude!! How are you so hyped every video
the fancy new import syntax didn't work with me. I tried using it (not in node).
could you do a 3d perlin noise for your next coding challenge
I love your work.. always! 🔥
Can I stop doing jumping jacks yet?
that's amazing #something
What IDE does he use does it work on Windows and Linux? For Eclipse is there a tensorflow.js plugin?
He is using Atom atom.io/
great!
Using tanh activation would bring the error rly low. I tried using it and in only 10 epochs the error went as low as 0.00002
I don't understand anything, but i love it. xd
go and get 1080ti and do lots lots processing to see the result
Neuroevolution with TensorflowJS soon? =)
What does For (i = 0, i < Bla, i++) do? (Random example)
#something
When I took out noLoop(); after I implemented model.fit() it never works. It says it can't start another fit() function even though I am using async/await
You need to ensure that the last fit is done running. Continue the video to see how he moves the rerunning of it to .then(() => {}). My guess is that in the version of tensor flow he is running there is no error output for the error (even though it is there). Google probably added the error later, to ensure fit is not run multiple times at once. :)
@@resin01 Thank you so much!
Was wondering if anyone had some advice, i am looking at entry level programming jobs and watching these tutorials although fun, is kind of discouraging. This guy is so fluid in his coding and i cant help but feel like crap thinking this is the level i need to be on, and i feel no where close.
You've stumbled on one of the less useful, more complex (and not necessarily in a good way) videos. Maybe try these playlists to start? You can do it!
czcams.com/users/shiffmanplaylists?view=50&shelf_id=14&sort=dd
czcams.com/users/shiffmanplaylists?view=50&shelf_id=2&sort=dd
On white board it reads as XOR wtf layers
this was helpful
Hey in the linear really regression challenge everything is working but stuck at the tf.train.sgd is giving error i have rechecked 100 of time but didn't solve the problem
Pulkit Rajpal what error
ShadowPlayz tf is not defined
Pulkit Rajpal well post the code here
Wow gitu lho..
is tensor flow for backend or client side?
I'm running tensorflow.js as a client side library here, but it can also run server side via node!
Daniel, could you make a game where there's no friction and the player never slows down, but it can accelerate in the direction you turn the player in. So for example, you turn the player 90 degrees around it's own center, and when you press the up arrow, the player moves towards that direction. I tried doing this but i'm really stuck, so any help would be appreciated!
Ese mi loco del código
I did #something.
PS: Great video Dan.
#something Great video! Keep up good work :D
Is there a playlist of coding challenges?
Here you go! czcams.com/play/PLRqwX-V7Uu6ZiZxtDDRCi6uhfTH4FilpH.html
The Coding Train Thank you!
Thank you
what is this text editor? sublime text ?
Atom. This workflow video might help:
czcams.com/video/HZ4D3wDRaec/video.html
Also:
sublime text: czcams.com/video/UCHzlUiDD10/video.html
atom editor: czcams.com/video/d3OcFexe9Ik/video.html
brackets: czcams.com/video/nmZbhManVcY/video.html
codepen: czcams.com/video/5gfUgNpS6kY/video.html
#Something
I love your tutorial so much!! Thank you >///
watch 50 shades of grey at 42:30 lol
nah i'll stick to programming buttons in banking apps xD
Hello Daniel, can you answer my questions(not coded related) on Quora?
#something :)
Thx a lot for all your videos. I don't look at all your live streams, but I try to look at all your edited videos and I learn a lot of thinks. Thx :)
Would it be possible to have a genetic algorithm with tf.js ? If so could you remake your flappy bird genetic AI with tf.js please ?
Love & gratitude.
I'd like to get to this eventually yes.
hastag something
what if failor instead of error but like
ah brb
#something 😅
12:15 😂😂
#Something!
BUT wait a second the XOR probleme can be solved at 100% sure with this formula : abs(X - Y)
Come on dude, switch to vs code already!
am i alone in being rather tired of TensorFlow? like is he being paid for this non stop stream of challenges using it?
James Wilson there's not that much tensorflow videos, he just enjoying it.
Also promises video series is still going I think.
the last 5 were all TensorFlow. and he's more than welcome to enjoy it, but I don't want to see him start to lose subs because he's not changing things up anymore. (not saying *I* would either) but it's something that happens a lot. people get comfortable with something and stop pushing themselves.
He ain't going to lose any subscribers which aren't that important, views are important which keep increasing, also TensorFlow is the new hotness and will be for quite a while, also it's a whole new field for him, of course it's hyper interested.
Basically, suck it up man, it's not going anywhere anytime soon.
Thanks for this feedback! I've heard the same from others and took a break from the ML stuff for today's stream. I will be focusing a lot on ML this summer, but it's not my plan for it to be the primary topic for the channel.
The Coding Train thank you for understanding my intentions as opposed to the over zealous fellow above lol
What a mess
#something
#something
#something
#something
#something
#something
#something
#something
#something
#something