ml5.js: Train Your Own Neural Network

[@TheCodingTrain]

🚂🌈🧠Code and community contributions! thecodingtrain.com/Courses/ml5-beginners-guide/6.1-ml5-train-your-own.html

[@planktonfun1]

you can add a replay function so you don't have to click again

[@masternobody1896]

Yeah machine learning thank you very much

[@pacoperico1109]

Hi! I really like your way to teach. It is brilliant. But, in neurals I have always doubts. For instances, Can I to use this kind of networks for predict a future result? Can I have an email for better explanation?

[@TheCodingTrain]

@@pacoperico1109 Come and join the coding train discord! thecodingtrain.com/discord

[@pacoperico1109]

@@TheCodingTrain hi again, I follow your advice and I am into discord. How do I find the coding train community?

[@ShreyAroraev3]

i really appreciate the fact that dan ALWAYS gives credit to original creators and researchers

[@Jptoutant]

No wonder he has so many friends, guy always doing a favor

[@sonmangaking]

He is a angel I wish I could work with him he inspires my work everyday

[@freeshavaacadooo1095]

Unlike a man known as Siraj Rival...

[@abeechr]

Shrey Arora agreed. And that’s why he’s closing in on a million subscribers and nary a negative comment.

[@ishanpradhan05]

Siraj Rawal triggered! xD

[@Cerzus]

6:02 Respect for flawlessly writing '0' and '1' while saying '1' and '2'.

[@TheCodingTrain]

😱

[@vikramb183]

I took a long hiatus from programming, but I didn't realize how long until I looked at the date this was posted! I'm finally getting around to finishing this after over a year! Also I'm so happy new generations of programmers will have such amazing videos to learn from!

[@williamlehmus5371]

You got me into programming! Please never stop doing these videos!

[@sonmangaking]

I love how you stopped and saw changes UR channel has become the golden light for teaching great topics like these I appreciate your work and would love to one day work or help. Your journey as it has shaped mine to becoming a great scientist

[@rkadi6540]

If my teacher is as cheerful as you when teaching, class wont be boring as hell. Man good presentation

[@grevel1376]

I made simmilar project about 2 weeks ago, and I am proud of myself, programm takes users clicks, left button makes red dots, middle button makes green dots and right button makes blue dots. Then a neural net trains on that input, and each 10 backprops it prints every pixel prediction, on rgb mix of confidences. It can be really beautiful, and it can find complex shapes. I made it in java awt.

[@tipsstudio9707]

can you give me link? so i can refer it.

[@grevel1376]

It is not published, maybe I'll make something towards it

[@tipsstudio9707]

@@grevel1376 ok dude, thanks i like to watch it when you published..

[@Akshatgiri]

There is no better teacher than Dan. Prove me wrong

[@shubhrashukla7879]

So close to 100k!!!!!!!!!!! Dan I love your vids and have been watching your channel from 160k from different accounts Thank you for turning me into a computer geek!

[@sarcasmasaservice]

Thanks Dan, this was great!

[@TheCodingTrain]

Yay, glad you liked it!

[@rsrini7]

All your videos are inherently number of views = number of likes. I don’t have a better way to appreciate your way of teaching. I love teaching. 🙏

[@exxzxxe]

Dan; you are a gentleman and scholar!

[@geoffwagner4935]

o-m-g south park The Toran Ra , with neural networking , he is the master

[@AkhilsaiNo1]

This world needs more people like Dan.

[@TylerMatthewHarris]

Your neural network is training my neural network to train a neural network. So Meta bro 🤪🤘🏻

[@divyeshgaur]

your enthusiasm made me excited to do ML. thank you and amazing video.

[@aarensan]

This could be a great start into ML ans JS - thank you!

[@jackal8004]

Oh my god, it's amazing! Really looking forward to the saving data part =)

[@prosoftwebindia]

Because of your the machine learning is easy for me.... big big thanks.. for nice 👍 contents.

[@ShreyAroraev3]

i was WAITING for this!!

[@IvanChernykh-vc1iv]

Absolutely mind blowing! Thanks Dan!!

[@SimonTiger]

We've not really only used pretrained models before. Instead, we did train our own models, but we always used a pretrained model as a basis. In this case, we're starting with a blank slate. No magic transfer learning or knn.

[@Clinty_Stwood]

I say "wow"...super good explanation and video realization

[@allex3842]

Thank you so much for everything you do ❤️

[@bafana_mhlamvu]

Greetings Mr. Shiffman, I noticed the Programming With Text series has been removed from your channel's playlist and was wondering if that was intended?

[@gamestv4875]

I wish I had the passion for coding like this guy. Recently I have no passion for anything at all. Scary.

[@hrrrvvyy]

Wow, it's amazing 😻

[@АнатолийП-я2н]

Hi, how to stop learning in whileTraining function depending on loss?

[@thomasgibson1028]

Aww my life I had to fight to understand normalization☺️. Thank you!

[@anteconfig5391]

22:44 you wrote '0.25' and I read '0.2J' and then I thought "what if I use complex numbers to calculate the weights and bias?" and stuff. I think I might try. I just came up with an idea to use a genetic algorithm that stores the starting state of the game of life and instead of doing a 2D space I use a 3D space.

[@Dron008]

Great demo!

[@fbn7766]

Love your videos............

[@micah7009]

Can the input be an array, or do you need to list each item in the array?

[@wtalkie]

Amazing as always!

[@mamupelu565]

this is getting good, go on please

[@mobsean]

So cool to watch this. Makes me wanna start coding RIGHT NOW !!! :-)

[@grainfrizz]

You're such a gift.

[@EdMonk]

That was fun. It worked for me.

[@jonathancoder]

Muy clara la explicación 👲

[@Manatoro]

duude you are too good!

[@kisame_5331]

Love your explanations!

[@vilkillian]

26:50 i don't get it, why don't you normalized data before model.classify?

[@mesutdogansoy]

because isNormalized flag set true when training so it normalizes the sample automatically

[@yamimzumara6291]

We need more, and your videos and awesome.

[@thisdot6205]

Could you make a snakeAI with reinforcement learning?

[@devorrahtester5439]

Thank you! Awesome tutorial! Would you explain how the training and output data would be structured if you used your hands to play the sound but also wanted to modulate the output sound with your legs and head?

[@satychary]

You are a nut, my kind :) Great video and project.

[@konstantinosvasilakos1240]

Is there a way to adjust the weights in the model using this example?

[@shashidhara.g.m655]

*YEEYYY ANOTHEERR VIDEOOO*

[@guyincognito1985]

Nice haircut! 👌😍👍 (If you're watching this ML playlist in order) I too hope to get a haircut one day and get rid of this COVID-mullet. 😟

[@scodemy1976]

Am a web developer and I have my channel but your videos are awesome

[@geoffwagner4935]

it also has me thinking multiwii pro , yes arduino and all of this. i still just don't know what my drone would be doing with all this yet. but objects and sensors, they have some pretty intelligent drones for synchronous flying, i wonder what kind of intellagent drone could be made with this stuff? it can see, this really beats a proxy sensor . it can know what things are , maybe to pick up and sort xD

[@nightshining5662]

This is a great overview! Anyone happen to know if there is a list of other open source ML 5 type libs for other languages like Python and C++? I want to run this type of stuff off google coral and Nvidia jetson without having to know TensorFlow

[@enivaldobonelli]

Great to start with 'tabula rasa.' Some artificial intelligence of previous codes did not work well for absence of human intelligence, in training the models. For example: a stretched hand shown to the program was identified as 'Band-aid!' Because some intelligent trainer showed her/his band-aid to the camera forgetting that more hand than band-aid were seen. Then if you make a fist, it is not a 'band-aid' anymore. Now it is a screw driver. Again, the trainer showed a small screw driver to the program, were mostly hand appeared.

[@WeirdSide]

can you use ml5.js to generate GAN images?

[@stiqsify]

So exciting

[@BarYamin]

Why not play all 3 notes in superposition based on their output probability according to the position of the mouse on the canvas?

[@micah7009]

I am getting an error when trying to train the data. "You are passing a target array of shape 91,1 while using a loss 'categorical_crossentropy'. 'categorical_crossentropy'expects targets to be binary matrices (1s and 0s) of shape [samples, classes]."

[@FrancoCervigni]

So what is loss? Me: That's a meme

[@thomasgibson1028]

Franco Cervigni 😁

[@sebastiendeseille9890]

Wonderful project ! Thanks !

[@codefun3168]

Great dan!!appreciate it..really inspired by you man!trying my best on my channel as well

[@codefun3168]

Need your help as well to improve coding channels on youtube..

[@codefun3168]

I find there are very less channels that promote coding.few are inactive..we need to bring BACK THE ENERGY!!

[@L76videos]

Can we read a folder to get images files from a browser?

[@abduljalilabdallah9578]

Hello everyone, I have a question, I need to run a Php code when my is condition is true in p5.js. unfortunately it doesn't work like in normal Js. If(number

[@vadsavin]

Is it possible to train models using "Teachable machine with google" not only to classify images, but to recogize where the objects are in the picture. So I want to get something like bounding boxes and classes of them for self-driving car with Raspberry Pi.

[@igotapochahontas]

Could you possibly do an offline version? A simple classification that does not require cloud services after being downloaded locally?

[@igotapochahontas]

Normally I just locally host the min.js file.... But I couldn't find that for ml5

[@hadiriazi]

What about classifying accelerator data (x,y,z) and determine standing, bowing, sitting ? Can I use the same approach? how about the data? how do I batch them ?

[@TheCodingTrain]

Yes, this could work! If you save your data to CSV or JSON you can load it directly into ml5. There are options to specify batch size and more, but ml5 will choose defaults for you.

[@hadiriazi]

The Coding Train great! Could you give me a head start or even better do a sample video on human activity recognition ? My main concern is how do I differentiate between bowing down and the reverse ... I just don’t know what would be each window’s data ?

[@TheCodingTrain]

You would have to label the data, i.e. record it somehow while you are also recording the classification (standing, sitting, etc.). The issue is that accelerometer data is 0 when you are still (or moving at a constant velocity). So it might work better for classifying a motion!

[@williamlehmus5371]

Have I found a bug or is ML5js supposed to work like this. If you add 10 data points and then check model before training you'll find them in: model.neuralNetworkData.data.raw If you then train the model, and check again, the data has been copied to model.data.training I was kind of expecting ML5js to move the data rather than copy them. In my understanding it's something like this: model.data.training = Data points that have been trained on. model.neuralNetworkData.data.raw = Data not trained on yet. I was expecting the data to be moved from raw to training, because if you now add more data to the model, the old data will still remain thus creating a weighting towards the old data since it's added twice.

[@ameynaik2743]

Great videos! Thank you. What is the default structure of the neural network in this case? I.e., number of hidden layers etc?

[@ameynaik2743]

learn.ml5js.org/#/reference/neural-network - never mind found it here. Thanks!

[@dimasandhikasury6128]

Nice to see you again.. :)))) can you make "NEAT" videos for your flappy brain??

[@DigitalAlligator]

model.addData() how much data it can hold?

[@rizkimramdani]

is this the same proccess if i want to train .jpg image file?

[@landonbrown2542]

I love this guy :)

[@liavofra]

Hello Mr coding Train, i have some math function that i am wondering about the geometry. the function is from a model i made, ill be happy to show you the model and hopefully you can help.

[@HoLeeFuk97]

how we train our image but they are in a folder? Sorry for my bad English

[@Akosiyawin]

Can I use 1000 images as an input? How?

[@8followsonik]

You can't play c sharp note, because it's JS. Kappa

[@Brahvim]

xD

[@TRT_MOOSIC]

Hey but I cant get a graph o'er there!!

[@CptUhudini]

I'm new to machine learning and I'm using ml5js for a current project. So I have a .csv file and an image folder with thousands of images (from @muxspace on GitHub). The .csv file maps every image in the image folder to a facial expression. I just don't understand what inputs and outputs I have to write in my options to train the model. Any ideas?

[@alexandrecsu]

How can I classify if I have inputs: "first_name", "last_name" and output: "gender"... what I am asking, what do I do if I have strings as imputs?

[@angelcaru]

function getNumberArrayFromString(str) { const result = []; for (const char of str) { const cc = char.charCodeAt(0); result.push(map(cc, 0, 255, 0, 1)); } return result; }

[@moonaddict]

Same question, did you find any solution?

[@PyroFlamingo]

Is anybody else having issues with the training? Dan's debug graph is so smooth but mine is all over the place and therefore when I test the prediction it is usually quite wrong.

[@ahmadooof]

that's so cool man

[@tawnyop7275]

I was following along with this and I legit spent like 20 mins trying to figure out why mine wasn't working. I just woke up and realised I didn't put the options into the neutral network smh

[@shubhrashukla7879]

Coding challenge: maybe you use the teachable machine or ML5 to classify a chord made by your hand and then you make it play the classified chord! That would be like a virtual ukelele

[@Zippka_]

is there something like this that I could use with Pyhton?

[@ReinaldoAssis]

Nino Orešković High level Tensorflow

[@selmirmuminovic551]

Thanks man!

[@aradhyamodi7761]

Thanks bro

[@gilberttheregular8553]

Finally!

[@moonaddict]

Oh man I'm so disappointed, I spent hours getting the data, organizing it, solving the training issues, and then when I went to classify... Boom: "expected dense_Dense1_input to have shape [null,90] but got array with shape [1,1]"... Apparently string inputs are not allowed? :(

[@moonaddict]

My dataset is a bunch of hexadecimals concatenated as a string, it's the data from a 3D map of a game, something like this: [{"map": "123456789", "changes": "geometry"}, {"map": "113556789", "changes": "particles"}] So, my goal was to get the label ("changes") identifying map changes based on its data as an input.

[@geoffwagner4935]

it would be fun to make an Arduino sorting drone, maybe collects certain shells along a big beach

[@johncerpa3782]

awesome vid

[@dmc5835]

Dan, can we get a a video on S.O.L.I.D principles?

[@harshaaddanki4606]

TF Lite plz

[@flamexode]

const coding = ml5.neuralNetwork(1, 1); coding.train(); If you don't get it you don't know what channel you are watching.

[@flamexode]

Dang, really appreciate how deep you go through the comments to get the best one

[@52.yusrilihsanadinatanegar79]

SyntaxError: ml5 is not defined

[@abhaykashyapa]

There is a hole in your laptop(green screen lol)

[@Gilotopia]

Is ml5js using the GPU?

[@TheCodingTrain]

Yes! It uses your local GPU via WEBGL. (This is not the work of ml5, but rather tensorflow.js underneath the hood.)

[@KnakuanaRka]

Idea for how to improve on this: instead of manually picking points to predict, once you have the training data, have it do predictions for the entire canvas, and color each pixel according to the prediction (say, C, D, and E being red, yellow, and blue), so we can more precisely see what the network predicts.

[@fabiano-co]

Man, I was playing with p5.js and the page just crashed, everything was gone and the page is blank, I lost my entire code. lol

[@TheCodingTrain]

I am so sorry to hear about this! It can happen with online editors like p5, I hope you aren't too discouraged! Try to save often :)

[@fabiano-co]

The same code, reducing the number of columns and removing those ones with zeros, worked very well.

[@prachetass]

Dn know if iam being rude but Match 3 game please........

[@ex1strides715]

I want you to be my professor!

[@thevoyageof67]

Sup Dan