Hands-On Machine Learning for Algorithmic Trading

MAP inference

We obtain point MAP estimates for the three parameters using the just defined model's .find_MAP() method:

with logistic_model:    map_estimate = pm.find_MAP()print_map(map_estimate)Intercept   -6.561862hours        0.040681educ         0.350390

PyMC3 solves the optimization problem of finding the posterior point with the highest density using the quasi-Newton Broyden-Fletcher-Goldfarb-Shanno (BFGS) algorithm, but offers several alternatives, which are provided by the sciPy library. The result is virtually identical to the corresponding statsmodels estimate (see the notebook for more information).