logistic-regression-pymc3.py

model = Model()

# Y = [1, 0, 0, 1, 1, ... 1, 0, 1]  # binary computed from n_predictors features

with model:
    # Priors for unknown model parameters
    # alpha = Beta('alpha', alpha=0.5, beta=0.5) 
    beta = Normal('beta', mu=0, tau=2. ** -2, shape=(1, n_predictors))
    
    # expected probability 
    p = tinvlogit(sum(beta * predictors, 1))
        
    # Likelihood (sampling distribution) of observations
    Y_obs = Bernoulli('Y_obs', p=p, observed=Y)
    

with model:
    try: 
        trace = load('trace.sqlite')
    except: 
        backend = SQLite('trace.sqlite')
        # obtain starting values via MAP
        start = find_MAP()

        # instantiate sampler
        step = NUTS(scaling=start)

        # draw posterior samples
        trace = sample(sample_size, step, start, trace=backend)