modifies q_next update in leapfrog integrator

[cablanc]

Hamiltonian-Generative-Networks/utilities/integrator.py

Line 131 in 702d3ff

q_next = q + p_next_half * self.delta_t

I think there is an issue with the leapfrog step. The current implementation updates q using:

p_next_half = p + dp_dt * (self.delta_t) / 2
q_next = q + p_next_half * self.delta_t

I think it should be changed to:

p_next_half = p + dp_dt * (self.delta_t) / 2
dq_dt, _ = self._get_grads(q, p_next_half, hnn)
q_next = q + dq_dt * self.delta_t