Episode 2 - Image Classification with Logistic Regression

YouTube Livestream: https://www.youtube.com/watch?v=a9HTZJMelv0

Notebooks used during this webinar:

Please use this thread to ask questions, and upvote questions you like. Thanks for joining!

1 Like

​how to do feature engg in pytorch fr a dataset? should we do it in python n then apply pytorch code for better results??


Go to https://jovian.ml/aakashns/03-logistic-regression and click the run button. If you are able to run the notebook, like this post.


What’s the below statement used for?
test_dataset = MNIST(root=‘data/’, train=False)

What does the np.random.permutation do?

Please explain more about nn.Module

Great presentation. thanks Aakash

Please spatial relationship

​why do we want to make an image into a batch of 1 image for prediction?

We do this because PyTorch models always require inputs to be fed in batches. It can’t accept a single training example. So when we want to predict on a single image, we pass it in as a batch of a single image.

okie… thank you! but i thought we were writing custom functions for everything - computing loss, training, validating/evaluating… am i missing something?

that’s right, but we used the built-in nn.Module class from PyTorch, which has this predefined behaviour of taking inputs as batches.

makes sense! thanks a lot!

So what is difference between keras and pytorch frameworks