fix infer_params and remove learning method

pymdp/jax/agent.py

@@ -254,51 +254,6 @@ def unique_multiactions(self):
         size = pymath.prod(self.num_controls)
         return jnp.unique(self.policies[:, 0], axis=0, size=size, fill_value=-1)
 
-    @vmap
-    def learning(self, beliefs_A, outcomes, actions, beliefs_B=None, lr_pA=1., lr_pB=1., **kwargs):
-        agent = self
-        if self.learn_A:
-            o_vec_seq = jtu.tree_map(lambda o, dim: nn.one_hot(o, dim), outcomes, self.num_obs)
-            qA = learning.update_obs_likelihood_dirichlet(self.pA, o_vec_seq, beliefs_A, self.A_dependencies, lr=lr_pA)
-            E_qA = jtu.tree_map(lambda x: maths.dirichlet_expected_value(x), qA)
-            agent = tree_at(lambda x: (x.A, x.pA), agent, (E_qA, qA))
-
-        if self.learn_B:
-            beliefs_B = beliefs_A if beliefs_B is None else beliefs_B
-            actions_seq = [actions[..., i] for i in range(actions.shape[-1])] # as many elements as there are control factors, where each element is a jnp.ndarray of shape (n_timesteps, )
-            assert beliefs_B[0].shape[0] == actions_seq[0].shape[0] + 1
-            actions_onehot = jtu.tree_map(lambda a, dim: nn.one_hot(a, dim, axis=-1), actions_seq, self.num_controls)
-            qB = learning.update_state_likelihood_dirichlet(self.pB, beliefs_B, actions_onehot, self.B_dependencies, lr=lr_pB)
-            E_qB = jtu.tree_map(lambda x: maths.dirichlet_expected_value(x), qB)
-
-            # if you have updated your beliefs about transitions, you need to re-compute the I matrix used for inductive inferenece
-            if self.use_inductive and self.H is not None:
-                I_updated = control.generate_I_matrix(self.H, E_qB, self.inductive_threshold, self.inductive_depth)
-            else:
-                I_updated = self.I
-
-            agent = tree_at(lambda x: (x.B, x.pB, x.I), agent, (E_qB, qB, I_updated))
-
-        # if self.learn_C:
-        #     self.qC = learning.update_C(self.C, *args, **kwargs)
-        #     self.C = jtu.tree_map(lambda x: maths.dirichlet_expected_value(x), self.qC)
-        # if self.learn_D:
-        #     self.qD = learning.update_D(self.D, *args, **kwargs)
-        #     self.D = jtu.tree_map(lambda x: maths.dirichlet_expected_value(x), self.qD)
-        # if self.learn_E:
-        #     self.qE = learning.update_E(self.E, *args, **kwargs)
-        #     self.E = maths.dirichlet_expected_value(self.qE)
-
-        # do stuff
-        # variables = ...
-        # parameters = ...
-        # varibles = {'A': jnp.ones(5)}
-
-        # agent = tree_at(lambda x: (x.A, x.pA, x.B, x.pB, x.I), self, (E_qA, qA, E_qB, qB, I_updated))
-
-        return agent
-
-
     @vmap
     def infer_parameters(self, beliefs_A, outcomes, actions, beliefs_B=None, lr_pA=1., lr_pB=1., **kwargs):
         agent = self
@@ -344,6 +299,8 @@ def infer_parameters(self, beliefs_A, outcomes, actions, beliefs_B=None, lr_pA=1
                 I_updated = self.I
 
             agent = tree_at(lambda x: (x.B, x.pB, x.I), agent, (E_qB, qB, I_updated))
+
+        return agent
 
     @vmap
     def infer_states(self, observations, empirical_prior, *, past_actions=None, qs_hist=None, mask=None):