Hands-On Machine Learning for Algorithmic Trading

Bayesian machine learning with Theano

PyMC3 was released in January 2017 to add Hamiltonian MC methods to the Metropolis-Hastings sampler that's used in PyMC2 (released in 2012). PyMC3 uses Theano as its computational backend for dynamic C compilation and automatic differentiation. Theano is a matrix-focused and GPU-enabled optimization library that was developed at Yoshua Bengio's Montreal Institute for Learning Algorithms (MILA) and inspired TensorFlow. MILA recently ceased to further develop Theano due to the success of newer deep learning libraries (see Chapter 16, Deep Learning for details). PyMC4, which is planned for 2019, will use TensorFlow instead, with presumably limited impact on the API.