Recurse Center - Batch 2 - Cycle 20241004-20241006 - Transposition


Transposition

This cycle was consumed by ARENA, plus work from the Neural Networks from Scratch book/videos. I did a lot of work with tensors, einops, linear algebra, and manipulating matrices. I'm still trying to build up an intuitive feeling for thinking and programming in this way, so I'm taking every mistep as a sign that I'm really working at the edge of my abilities - whenever I have to learn something new, I take it as evidence of that fact. And I learned a lot of new things.

Day 1

Today I started the exercises for ray tracing

ray tracing

torch.linspace

torch.linalg.solve

torch.stack

torch.transpose

torch.eye

I was also able to pair with another Recurse and get onto the GPU machines on Heap :)

Day 2

I worked through Neural Networks from Scratch (NNFS) Chapters 1 and 2

I also learned more about PyTorch for manipulating tensors, matricies, vectors, arrays...

torch

torch.cat

vs.

torch.stack

Day 3:

Today I finished the ray tracing exercises, with some help from an AI assistant (I use Claude Sonnet 3.5). The exercises actually recommended we do that, and it was nice to work with something (I almost wrote someone) that I can share ideas and approaches with, confirm my intuition, and help lead me to the right implementation. This was super helpful, because these last exercises were pretty chanllenging for me. We dealt with 2D rays and seeing if they interset with triangles. All of this was meant to help build up to a function that rendered a mesh of Pikachu.

Some more PyTorch functions that I came across:

torch.linspace

torch.reshape

torch.linalg.solve

torch.Tensor.expand

torch.squeeze

I also came up against broadcasting, which I need to spend more time internalizing.

Lastly I finished NNFS Ch 3

Things for next cycle