store log_lik_t in attribute

pymc3_hmm/step_methods.py

@@ -146,6 +146,7 @@ def __init__(self, var, values=None, model=None):
 
         self.gamma_0_fn = model.fn(var.distribution.gamma_0)
         self.Gammas_fn = model.fn(var.distribution.Gammas)
+        self.log_lik_t = None
 
         super().__init__([var], [dependents_log_lik], allvars=True)
 
@@ -160,8 +161,10 @@ def astep(self, point, log_lik_fn, inputs):
         # could be missing out on a much more efficient/faster approach to this
         # potentially large computation.
         # state_seqs = np.broadcast_to(np.arange(M, dtype=int)[..., None], (M, N))
-        # log_lik_t = log_lik_fn(state_seqs)
-        log_lik_t = np.stack([log_lik_fn(np.broadcast_to(m, N)) for m in range(M)])
+        if self.log_lik_t is not None :
+            log_lik_t = self.log_lik_t
+        else:
+            log_lik_t = np.stack([log_lik_fn(np.broadcast_to(m, N)) for m in range(M)])
 
         return ffbs_astep(gamma_0, Gammas_t, log_lik_t)