So well come to my guide for HDC(hyperdimension computing)for beginners and dummies, so ur here to learn HDC good, very maso, love it, 1.first u need python(1.0+)
2.next for libraries it's pytorch + torchdd or holo(If ur feeling extra kinky) which is a rust-python wrapper,
- AI example code import torch import torchhd
1. Set the dimensionality
d = 1000 # The "1000D" you mentioned
2. Create a random hypervector (your first "memory")
hv = torchhd.random(1, d) # A tensor with 1,000 random values
print(f"Hypervector shape: {hv.shape}")
3. The fundamental operations
hv2 = torchhd.random(1, d)
BUNDLE (superposition): Think of this as creating a "set" or a "noisy" memory
bundled_hv = torchhd.bundle(hv, hv2)
BIND (association): This creates a new hypervector that encodes a relationship
bound_hv = torchhd.bind(hv, hv2)
4. Similarity (the "search" and "recognition" function)
similarity = torchhd.cosine_similarity(hv, hv2)
print(f"Similarity between two random vectors: {similarity.item():.4f}")
You'll see a number close to 0, as random vectors are nearly orthogonal.
(I recommend claude for free and gemini pro if ur rich 🤑to code for you if ur feeling extra romantic)
- I recommend starting at 1k D, cos it's easier to control and won't get you scratching ur head but itching for more
- I recommend thinking them of a sheet of graphene(2D) to help u visualise, if ur feeling smart try thinking them of diamond lattice(3D), if u feel even spicier image them as a universe with galaxies as moving nodes cos "distance" is now an abstract concept
- Stay hydrated
- Have fun 😊
- Ask questions, cos it's better if ur responsive and ready to learn -teehee ✨😘
Top comments (0)