// the find
goodfeli/theano_exercises
Exercises for my tutorials on Theano
Tutorial exercises for Theano, the Python numerical computation library that predates TensorFlow and PyTorch. Written by Ian Goodfellow (of GAN fame) around 2014–2016, covering symbolic differentiation, GPU execution, and basic ML models. Strictly historical at this point — Theano itself has been dead since 2017.
The exercise-plus-solution file pairing is a good teaching structure — you can attempt the exercise and check your work without switching documents. Covers non-obvious Theano concepts like symbolic graph traversal and two-step backprop that most tutorials skipped. Debugging section (NaN detection, compute_test_value) addresses real pain points that cost people hours.
Theano is abandoned and incompatible with anything modern — Python 3.x support was bolted on and the library is not pip-installable on current environments without significant pain. No chapter 3 in the directory tree (jumps from 01 to 02 to 04), so the curriculum has a visible hole. The README is four lines pointing to Google Drive slide links that may or may not still work. Zero value if you're doing any ML work today with PyTorch or JAX.