For great start, we should use some good tool for manuage the python and virtual environment We will be using UV. On your MacBook, we use a simple command line.

1

brew install uv

Then we make a new directory for contain our sourcecode.

1
2

mkdir alpha_fold
cd alpha_fold

We init our directory and add a few packages to make our own modal.

1
2
3

# in the same directory
uv init .
uv add numpy matplotlib torch torchvison jupterlab

From now on, we can actually focus on the building process without worry python virtual environment

Introduction

In this Introduction, I will give you some fundimental concept and keywords just to get used to machine learning framework and how we combine with mathmatical theory.

Tensor

What is a Tensor? This can be different based on different backgroud. In the field of AI, we use Tensor as a continous memory of the data. We use it to store the numerical data and compute with it.

Here in Pytorch framework, we can easily create one.

1

a: Tensor = torch.Tensor([1, 2])

Tensor has different dimensions, with can be used as a features of the data, since the protein structure is a 3d structure.

1

a.dim

The dimension of the Tensor, from outer to inner is : 0...N, from inner to outer is : -1..-N-1

Tensor operation obey the rule of matrix operation and vector operation

  1. Indexing and Slicing
  2. Operation
  3. broadcasting
  4. auto-derivatitive

Machine Learning

Machine Learning is about learning the pattern

  1. Similarity Estimation
  2. Classification

Loss Function: $ Loss(w) $ Chain rule AffineLine

Elegent torch.einsum()

Attention is all you need

Sequence of data

Attention: Neural Network

Handle the Sequence divide sequence of the words

RNN and Attention Machecism (watch build a large language model from scratch) NLP computer Vision