ddbourgin/numpy-ml ? reverse-engineered prompt

Build me a Python library for machine learning that’s written mostly in NumPy, with clear code and simple docs so it’s easy to learn from and tinker with. I want it to include the usual basics like linear and logistic regression, decision trees, random forests, gradient boosted trees, k nearest neighbors, naive Bayes, and clustering with Gaussian mixtures, plus a few sequence and neural net pieces like hidden Markov models, simple RNNs, LSTMs, attention, word embeddings, and a small variational autoencoder and GAN example.

It should also have practical helpers for preprocessing text and signals, distance and similarity utilities, and a few reinforcement learning and bandit examples. Please organize it as a proper installable package, add tests, and make the documentation easy to browse with short examples for each model. Keep everything self contained and readable, and use only standard Python plus NumPy unless there is a really good reason not to. If you need to look up anything current, feel free to check the docs online.