miCPPgrad

miCPPgrad (my c++ grad) is a tiny scalar-valued autograd engine in c++ inspired by Kaparthy's micrograd

in src/main.cpp, i wrote a sample MLP for 3D surface regression task given some noisily generated points from a ground_truth function

you can train it with

./run.sh

run testcases for the Value engine

./run.sh test

skip the build by adding skipbuild as an arg

./run.sh skipbuild

after running ./run.sh, you can generate a visualization gif of the training process with python graph.py

please make sure you have matplotlib and pandas installed in your environment to run the graphing script

here's a sample i made, trained model surface on the left and ground truth on right with noisy data on both sides