Add local search sampler

src/gfn/containers/trajectories.py

@@ -451,26 +451,94 @@ def reverse_backward_trajectories(trajectories: Trajectories) -> Trajectories:
                 )
             )
 
-        new_when_is_done = trajectories.when_is_done + 1
-        new_states = trajectories.env.sf.repeat(
-            new_when_is_done.max() + 1, len(trajectories), 1
+        # Compute sequence lengths and maximum length
+        seq_lengths = trajectories.when_is_done  # shape (n_trajectories,)
+        max_len = seq_lengths.max().item()
+
+        # Get actions and states
+        actions = (
+            trajectories.actions.tensor
+        )  # shape (max_len, n_trajectories *action_dim)
+        states = (
+            trajectories.states.tensor
+        )  # shape (max_len + 1, n_trajectories, *state_dim)
+
+        # Initialize new actions and states
+        new_actions = torch.full(
+            (max_len + 1, len(trajectories), *trajectories.actions.action_shape), -1
+        ).to(
+            actions
+        )  # shape (max_len + 1, n_trajectories, *action_dim)
+        new_states = trajectories.env.sf.repeat(max_len + 2, len(trajectories), 1).to(
+            states
+        )  # shape (max_len + 2, n_trajectories, *state_dim)
+
+        # Create helper indices and masks
+        idx = (
+            torch.arange(max_len)
+            .unsqueeze(1)
+            .expand(-1, len(trajectories))
+            .to(seq_lengths)
         )
-
-        # FIXME: Can we vectorize this?
-        # FIXME: Also, loop over batch or sequence?
-        for i in range(len(trajectories)):
-            new_actions[trajectories.when_is_done[i], i] = (
-                trajectories.env.n_actions - 1
-            )
-            new_actions[
-                : trajectories.when_is_done[i], i
-            ] = trajectories.actions.tensor[: trajectories.when_is_done[i], i].flip(0)
-
-            new_states[
-                : trajectories.when_is_done[i] + 1, i
-            ] = trajectories.states.tensor[: trajectories.when_is_done[i] + 1, i].flip(
-                0
-            )
+        rev_idx = seq_lengths - 1 - idx  # shape (max_len, n_trajectories)
+        mask = rev_idx >= 0  # shape (max_len, n_trajectories)
+        rev_idx[:, 1:] += seq_lengths.cumsum(0)[:-1]
+
+        # Transpose for easier indexing
+        actions = actions.transpose(
+            0, 1
+        )  # shape (n_trajectories, max_len, *action_dim)
+        new_actions = new_actions.transpose(
+            0, 1
+        )  # shape (n_trajectories, max_len + 1, *action_dim)
+        states = states.transpose(
+            0, 1
+        )  # shape (n_trajectories, max_len + 1, *state_dim)
+        new_states = new_states.transpose(
+            0, 1
+        )  # shape (n_trajectories, max_len + 2, *state_dim)
+        rev_idx = rev_idx.transpose(0, 1)
+        mask = mask.transpose(0, 1)
+
+        # Assign reversed actions to new_actions
+        new_actions[:, :-1][mask] = actions[mask][rev_idx[mask]]
+        new_actions[torch.arange(len(trajectories)), seq_lengths] = (
+            trajectories.env.n_actions - 1
+        )  # FIXME: This can be problematic if action_dim != 1 (e.g. continuous actions)
+
+        # Assign reversed states to new_states
+        assert torch.all(states[:, -1] == trajectories.env.s0), "Last state must be s0"
+        new_states[:, 0] = trajectories.env.s0
+        new_states[:, 1:-1][mask] = states[:, :-1][mask][rev_idx[mask]]
+
+        # Transpose back
+        new_actions = new_actions.transpose(
+            0, 1
+        )  # shape (max_len + 1, n_trajectories, *action_dim)
+        new_states = new_states.transpose(
+            0, 1
+        )  # shape (max_len + 2, n_trajectories, *state_dim)
+
+        # TODO: Add below into the test suite to ensure correctness
+        # new_actions2 = torch.full((max_len + 1, len(trajectories), *trajectories.actions.action_shape), -1).to(actions)
+        # new_states2 = trajectories.env.sf.repeat(max_len + 2, len(trajectories), 1).to(states)  # shape (max_len + 2, n_trajectories, *state_dim)
+
+        # for i in range(len(trajectories)):
+        #     new_actions2[trajectories.when_is_done[i], i] = (
+        #         trajectories.env.n_actions - 1
+        #     )
+        #     new_actions2[
+        #         : trajectories.when_is_done[i], i
+        #     ] = trajectories.actions.tensor[: trajectories.when_is_done[i], i].flip(0)
+
+        #     new_states2[
+        #         : trajectories.when_is_done[i] + 1, i
+        #     ] = trajectories.states.tensor[: trajectories.when_is_done[i] + 1, i].flip(
+        #         0
+        #     )
+
+        # assert torch.all(new_actions == new_actions2)
+        # assert torch.all(new_states == new_states2)
 
         trajectories_states = trajectories.env.states_from_tensor(new_states)
         trajectories_actions = trajectories.env.actions_from_tensor(new_actions)
@@ -480,7 +548,7 @@ def reverse_backward_trajectories(trajectories: Trajectories) -> Trajectories:
             states=trajectories_states,
             conditioning=trajectories.conditioning,
             actions=trajectories_actions,
-            when_is_done=new_when_is_done,
+            when_is_done=trajectories.when_is_done + 1,
             is_backward=False,
             log_rewards=trajectories.log_rewards,
             log_probs=None,  # We can't simply pass the trajectories.log_probs