Trainer.py

from scipy import optimize

class Trainer:
    def __init__(self, N):
        self.N = N

    def cost_function_wrapper(self, params, X, y):
        self.N.set_params(params)
        cost = self.N.cost_function(X, y)
        grad = self.N.compute_gradient(X, y)

        return cost, grad

    def callback_f(self, params):
        self.N.set_params(params)
        self.J.append(self.N.cost_function(self.X, self.y))

    def train(self, X, y):
        self.X = X
        self.y = y

        self.J = []
        params0 = self.N.get_params()
        options = {'maxiter':200, 'disp':True}
        res = optimize.minimize(self.cost_function_wrapper, params0, jac=True,
                                method='BFGS',
                                args=(X, y),
                                options=options,
                                callback=self.callback_f)

        self.N.set_params(res.x)
        self.optimization_results = res