rrtucci / texnn

Python script that genetates LaTex code that draws a Neural Net as a causal DAG (Bayesian Network). Python wrapper for xy-pic LaTeX package.

texnn (pronounced like "Texan") is a Python script that generates LaTex (tex) code that draws a Neural Net (nn) as a causal DAG (Bayesian Network).

texnn is capable of reproducing with ease most of the xy-pic generated bnets (Bayesian Networks) displayed in my book Bayesuvius.

texnn is a stand-alone app. I wrote it specifically to aid me in writing a chapter on transformer architectures (see directory "vanilla_transformer") for my free, open source book Bayesuvius. But I soon realized that it could be easily converted into a general tool that is independent of the transformer topic and of Bayesuvius.

texnn uses the LaTex package xy-pic for drawing. In broad terms, texnn can be described as a Python wrapper for the LaTeX package xy-pic.

Gallery:

• Asia

• Expert Archer

• Gaussian Markov random field (by Kevin P. Murphy)

• Rock Layers

• SentenceAx

• Silly Bnet

• Thermodynamics

• Vanilla Transformer

  • Scaled dot product attention
  • Multi-head Attention
  • Encoder
  • Decoder