Multi class Logistic regression implementation from scratch in python on MNIST dataset

Hey everyone,

This video is a walkthrough tutorial of multi class logistic regression in python which is a supervised machine learning task . Multi class logistic regression, also known as multinomial and multivariate classification or regression where the goal is to identify different objects into more than 2 classes e.g. a given image is of an apple orange or banana?
This video covers the implementation of this algorithm from start to finish in python without using any libraries like Scikit learn or Keras.

Classifier will be trained and tested on famous MNIST dataset which consists of images of hand written digits from 0 to 9. We will work with the pixel intensity values of each image and from that we will try to classify what digit the given picture is of. To make this work for the images of all the digits we will be using the strategy called one vs all, details of which are discussed in depth in the video.

Finally we will be checking the accuracy of our classifier that how good is it performing, which is very well in fact.

If this video proved any valuable then do give this video a thumbs up and dont forget to subscribe too, you definitely wouldn't want to miss the cool videos which are on their way ;)

There's a slight repetition of clip where Im explaining about the dataset which somehow managed to sneak past the final edit phase, so skip that

Time Stamps
intro 0:00 - 1:10
Theory background 1:10
Pseudocode 12:43
Implementation 14:52

#machinelearning #classification #python #tutorial #code #programming #walkthrough #multiclass

[joserivera]

everything was well said, simple and easy to understand, thank you

[hasnainkhawaja]

Very well explained! Couldn't find relevant material on the internet, they were all about how to use pre-built libraries

[nilupulperera]

Wow! Great explanation. This video helped me to understand the basics very well. Thank you very much.

[hamzazaman]

Very helpful and well explained. U rock keep up the good work 👍🏻

[kk]

Can you please tell me a unique problem statement which can be used for this

[donzhu]

Got error, can you help to explain it?
z = np.dot(weights, X) + bias
<_array_function_ internals> in dot(*args, **kwargs)
ValueError: shapes (1, 400) and (399, 5000) not aligned: 400 (dim 1) != 399 (dim 0)

[satkanova]

Hello, mistake : dict object has one attribute “at” in code

[harishh.s]

Hi,

Thanks for sharing your knowledge. I just had a doubt.For training the models you have considered 400 weights, whereas there are only 399 features (pixels) in X. I did not understand how the dot product between matrices (1, 400) and (399, 5000) did not give an error. This is the error I got when I ran the code. Can you please explain this?