Perceptron in Python - Machine Learning From Scratch 06 - Python Tutorial

[@architech5940]

This is honestly a great playlist. I've referenced this playlist a couple of times to practice these algorithms the same way you would practice solving and applying leetcode algorithms. I hope you do a series on deep learning such as DQN, RNN, and LSTM model architecture in the future.

[@snandhininvrzmsywqh2687]

love you max, have taken so many courses but none taught me the algo implementation from scratch. thanks a ton

[@patloeber]

thanks :) glad you like it

[@CallBlofD]

Hi, great video, really enjoyed it. For people coming only with a theory background, the version of the gradient descent he's using in the video is SGD (Stochastic Gradient Descent). Overall great video, thank you!

[@thealphadubstep5403]

If you question why you don't understand some parts of this video, it's because you've missed out on some parts of the theory. It was well explained and implemented. Thank you for giving me some more knowledge!

[@thealphadubstep5403]

​@Justin Williams Since my message was sent about a year ago, let me re-state what I said from back then. The video is an implementation tutorial. If you want to understand the theory behind a perceptron, a simple google search is more than enough. Though there are some interesting details that cannot be explained easily, while implemented they are already well explained on the web. Try checking out the article from medium.com Perceptron: Theory and Practice. Any other questions you might have, feel free to ask in the comments. I've implemented it myself not from this video but by understanding the theory behind it. Keep in mind, I might be a bit rusty on this project since it has been over a year now.

[@sidesplitter9497]

just wanna say a big thank you for all these ML videos, they're great! Any chance you could expand this to a multi-layered perceptron? I really like your style of coding and I think it would help to understand stuff like backprop. Thanks!

[@patloeber]

Thank you! Multilayer Perceptron is indeed on my list in the near future!

[@mfurquimdev]

I'm really enjoying your Machine Learning from Scratch series. It has been really useful for my machine learning classes. Thank you!

[@Slisus]

Hey Patrick, I'm trying to get into ML, that's why a stumbled over this ML from Scratch playlist. Thanks a lot so far for all the high quality content. There was just a question popping into my mind in this video, but also in previous videos as well: Why don't you call the predict() function inside the fit() function. It seems like you write the same code, you already wrote in the predict-function, again in the fit-function.

[@alexfrank1831]

Awesome, best ML Video i have seen so far!

[@arundas7760]

Great work, thank you sir I have a doubt on how bias got updated

[@vladislavkurichenko5413]

Thank you for your video! But can you please explain the logic behind this expression: x1_1 = -(p.weights[0] * x0_1 + p.bias) / p.weights[1]?

[@Дмитрий-с7х9р]

Great explanation, thanks!

[@nadabu.274]

Wonderful explanation. I read many resources but your video saves a lot of time. thank you. Can you explain backpropagation and RBF networks in the future?

[@patloeber]

Thank you! I already have a tutorial on backpropagation in my pytorch playlist. but I plan to do a neural net from scratch in the future

[@marilyncancino4875]

Thanks a lot for such an awesome video

[@patloeber]

Glad you liked it

[@fettahyldz460]

Why are we incrementing bias in every for loop instead of setting it to one at first? Thanks, great video.

[@patloeber]

Because "update" can also be negative, so we either increment or decrement here depending on our predictions...We have to find the bias through the training and cannot just set it

[@robosergTV]

@@patloeber why not pack the bias into the feature space so that you only have to do the dot product? Easier on the code and readability. Ofc the bias would be set to 1.

[@VidhiShah]

Where in the program are you defining "y_true" and "y_pred" for it to pick up in the accuracy function? def accuracy(y_true, y_pred): accuracy = np.sum(y_true == y_pred) / len(y_true) return accuracy

[@devfordevs]

we are passing y_test as y_ture and prediction as y_pred , we are not defining y_true seperately it is just varibale name .

[@charleskentucky6323]

Hi Patrick, Thanks for the videos. I was just wondering, how could i use the output of a single perceptron in your code as an input for other perceptron? As a context, i'm currently making a XOR gate which require AND, NOT, & OR gate combined into XOR gate.

i'm sorry for asking because i'm new in learning programming. may i ask where did the training data set came from for the perceptron to work on?'m always lost on where did the perceptron actually depend its dataset?

[@inscseeker401]

From the random dataset generated by sklearn make blobs I think

[@dhruvvaidh8843]

why are we converting the target variables into ones and zeroes

[@frederikpedersen4396]

First of all, thanks a ton for the great video! Really helped me a ton as supplementary materiel to my Scientific Programming class! Oh, and just a little heads up: the "v" in "activation" is pronounced as the 'v' in 'voodoo', 'valve' or 'viking' and not as the 'w' in 'wood', 'weekend' or 'weed'. Thanks again!

[@patloeber]

Thanks for the feedback and the tip :)

[@dhruvvaidh8843]

How do we apply this to IRIS as it has target values of 0,1,2

[@MahdiMashayekhi]

This video is great!❤❤

[@gitasaheru2386]

MultiLayer Perceptron

[@cherszardo9932]

Hi, I tried your code but I encountered the error “AttributeError: 'Perceptron' object has no attribute '_unit_step_func'” can u please help me. Thanks

[@dhruvlm]

make sure you use the self keyword whenever you're calling a function inside a class, for example if you're calling the unit step function, write self._unit_step_func(x)

[@patloeber]

yes that's right. You can compare your code with mine on GitHub

[@amanullahshareef1768]

What is Y in update rule explanation?

[@korypom]

Hi, why is your algorithm accuracy still pretty high despite initializing weight as 0? From what I know, weight should not be initialized as 0. Below is my implementation of the update rule and it works : if y_[idx]*np.matmul(self.weights, x_i)

[@patloeber]

I'm not seeing the learning rate in this formula. Do you apply it somewhere?

[@MrLanzudoo]

Hi Python Engineer. I was trying your code with a dataset of SMS ham/spam. One question: in your case the labels are inter (0 and 1)? Its because with my dataset it gives me an error "TypeError: '>' not supported between instances of 'str' and 'int'). Thanks in advance!!

[@patloeber]

your y should be integer class labels...looks like you have string types

[@valentinamovilsandoval1819]

I didn't understand why you make all the y 1 or 0 depending if they are greater than 0 or not.

[@wnurulain]

Can anyone share how to forecast tourist arrivals using MLP? Thank you

[@mohamedahmedmohamedzainada2461]

please cod for Perceptron learning : The Logical Operation XOR

[@xirog1]

Hello there :) . I run the code but it says "TypeError: fit() missing 1 required positional argument: 'y' " . Any idea why that happened?

[@patloeber]

you either defined fit not corretly (def fit(self, x, y)), or you don't pass all arguments to it

[@SuperOnlyP]

I have two questions: 1- Could you please explain what is: y_ = np.array([1 if i > 0 else 0 for i in y ]) - line 19 - Perceptron Class. Is it for label for the y_train . I thought the y_train datasets has its own label already. 2- for the update weight - line 26 - Perceptron Class. Does this line will create: 1 or -1 for updating the weight to be added or subtracted. Thanks !!!!!

[@patloeber]

1. Sometimes datasets have +1and -1, but in our implementation we need it as +1 and 0 because of how our update rule is designed. 2. It will create +1 and -1 for wrong classification, and 0 for correct classification. that's why 1) was necessary: y_ = 0, y_p = 1 --> (y_[idx] - y_predicted) = -1 y_ = 1, y_p = 0 --> (y_[idx] - y_predicted) = +1 y_ = 0, y_p = 0 --> (y_[idx] - y_predicted) = 0 y_ = 1, y_p = 1 --> (y_[idx] - y_predicted) = 0

[@SuperOnlyP]

@@patloeber thank you very much !

[@adamspackman6275]

Been enjoying these videos and they are really helping me with my course right now but I have a question. I changed some of the variables in the created data set and whatever else to test the accuracy function. Now I had my plot show the line through one of the big clusters so it definitely was not 100% accurate yet the accuracy function still returned with 1.0. I have double checked the code and they are 1 for 1 so I am just wondering whether this accuracy function is viable to use for this model?

[@adamspackman6275]

Or is it because some datasets cannot be linearly split in half so it is tricking the model into thinking it is correct when it is not? Although the random dataset I generated was able to be split in half with a linear line?

[@machinelearning4376]

Positive Aufgabe

[@luantorres3758]

hey man, can you please provide the formula's link too?

[@cristinaramos293]

Hello, when I run the code I get None as the accuracy score...any guesses why this is happening?

[@patloeber]

Hmm you can compare the code with mine on GitHub to find possible errors

[@anuragsharma4435]

what to do if i have to apply sigmoidal function and have 3 classes ?

[@patloeber]

You can’t use sigmoid for multiclass problems. You need to use softmax function in this case

[@magelauditore333]

whats the difference between logistic regression and perceptron

[@patloeber]

lr uses sigmoid function and different update rules. also have a look here: stats.stackexchange.com/questions/162257/whats-the-difference-between-logistic-regression-and-perceptron

[@samarthraizada]

I get a module not found error when i try ' from perceptron import Perceptron'. Please help!

[@patloeber]

try 'from . import Perceptron', or try 'from .perceptron import Perceptron'

[@Wolfgang7913]

what compiler are you using?>!?!?!!?

[@patloeber]

Python 3.7

[@_PrajaktaMokale]

Thanku 🥰🥰

[@luthfiarsyad358]

any solution if im using jupyter notebook? i need to teach my class with it :( welp guys

[@mahmoudabbasi1994]

Great!

[@VV-xt7fj]

These are great tutorials but I don't understand the purpose of learning rate here. Percepton is guaranteed to converge if the data set is linearly separable.

[@drishalballaney]

Where is the playlist?

[@patloeber]

kzbin.info/www/bejne/pJivqotracpkgrc

[@drishalballaney]

@@patloeber thank you very much, your explanations are really good

[@saurabhpatil4576]

11:05

[@sherlockmania749]

hey do you help me for my assignment

[@yantisuartini1855]

What the name of dataset ?

[@patloeber]

datasets.make_blobs(...). this is just a toy dataset that can be generated with sklearn

[@rezavelayani8223]

hy, i need flowchart from your code ? Thanks

[@puneetgupta1246]

lk

[@sherlockmania749]

please

[@meghaagarwal6071]

The voice is a bit irritating :|

[@alempongen5919]

Then don't watch lol

[@joseph3036]

Then stop watching!