How qkchambers/trabalho1-cps841 works
Project type and purpose
• This is an academic/research monorepo containing two related model implementations (BTHOWeN and ULEEN) of the Bloom WiSARD neural‐like classifier.
• Each implementation has two halves:
a. A hardware description (RTL) generator and testbench for SystemVerilog (via Mako templates).
b. A software reference model in Python for loading data, training, and evaluation.
• The goal is to reproduce and extend Bloom WiSARD results on multiple UCI datasets using both hardware‐mappable RTL and a software model.
File structure
• Top level: Docker/DevContainer configs, .env, README.md, and two sibling directories BTHOWeN/ and ULEEN/.
• Each of BTHOWeN/ and ULEEN/ contains:
• rtl/ – Mako templates (mako_srcs) + build scripts (rtl/build) + generated RTL + synthesis reports + testbench code.
• software_model/ – Python modules for data conversion, model definition, training, evaluation, and utility scripts.
• requirements.txt, README.md, and Makefile or setup.py for packaging.
• There's also a top‐level models/ directory (to store persisted model files) and global requirements.txt for Docker builds.
README for additional insights
• The README clarifies launch configurations (train_swept_models.py, evaluate.py, full_run.py) and the datasets used.
• It points out how example.json configures ULEEN parameters.
• It references academic papers and GitHub repos for the underlying algorithms.