Welcome to the MNIST-Tutorial repository! 🚀 This is your one-stop shop for playing around with the classic MNIST dataset in some creative ways. We've got everything from traditional CV methods to a multi-view approach. Dive in and enjoy! 😄

Here's the modified README with the new file kan_train_seq.ipynb and updated results table:

• get_data.ipynb 🕵️‍♂️
  Start here! This notebook is your data detective, exploring the MNIST dataset to uncover all its secrets. Perfect for getting to know your data better.

• train.csv 📊
  The heart of the project! This file contains the MNIST dataset stored as sequences. Each sequence has a length of 784, representing the flattened 28x28 grayscale images.

• train.ipynb 🖼️
  Feeling traditional? This notebook transforms the 784-length sequences back into 28x28 images and trains a model using classic computer vision techniques. Because sometimes, old school is the best school!

• train_seq.ipynb 📈
  Why complicate things? This notebook skips the reshaping and trains directly on the original sequences. Simple, yet effective!

• kan_train_seq.ipynb 🔄
  A new perspective on the sequence approach! This notebook introduces a different way to train directly on the original sequences, trying to squeeze a bit more out of the simple approach.

• train_multi_branch.ipynb 🔍🔬
  Can't decide between image and sequence? Why not both?! This notebook uses a multi-view approach, combining the power of both perspectives to train a model. Two heads (or views) are better than one!

Here's how our different methods stack up on validation accuracy:

As you can see, the multi-view approach gives us the best results! 🏆 However, each method brings something valuable to the table.

Clone the repo, open up the notebooks, and start experimenting with the MNIST dataset in these different ways. Whether you're a fan of images, sequences, or both, there's something here for you! 💻🎉