add function for generating TrajectoryContextDataset from a list of trajectories

kooplearn/data.py

@@ -206,10 +206,59 @@ def _contexts_from_traj_np(trajectory, context_length, time_lag):
     data = np.moveaxis(data, -1, 1)[:, ::time_lag, ...]
     return data, TensorContextDataset(idx_map)
 
+def get_multi_trajectorycontextdataset(
+    trajectories: ArrayLike,
+    context_length: int = 2,
+    time_lag: int = 1,
+    backend: str = "auto",
+    **backend_kw,
+    ):        
+        """
+        Generates a TrajectoryContextDataset from multiple trajectories.
+        It takes as input a list of trajectories and returns a sequence of context windows.
+
+        Args:
+            trajectories (ArrayLike): A trajectory of shape ``(n_trajs, n_frames, *features_shape)``.
+            context_length (int, optional): Length of the context window. Default to ``2``.
+            time_lag (int, optional): Time lag, i.e. stride, between successive context windows. Default to ``1``.
+            backend (str, optional): Specifies the backend to be used (``'numpy'``, ``'torch'``). If set to ``'auto'``, will use the same backend of the trajectory. Default to ``'auto'``.
+            backend_kw (dict, optional): Keyword arguments to pass to the backend. For example, if ``'torch'``, it is possible to specify the device of the tensor.
 
-class MultiTrajectoryContextDataset(TrajectoryContextDataset):
-    def __init__(self):
-        raise NotImplementedError
+        Returns:
+            data (TrajectoryContextDataset): A sequence of context windows aggregating all trajectories
+        """
+
+        if trajectories.ndim < 3:
+            raise ShapeError(
+                f"Invalid trajectories shape {trajectories.shape}. The trajectories should be at least 3D."
+            )
+        torch, backend = parse_backend(backend)
+
+        #initialize TrajectoryContextDataset with the first trajectory
+        data = traj_to_contexts(trajectories[0],context_window_len=context_length,time_lag=time_lag,backend=backend,**backend_kw)
+
+        if torch is not None and torch.is_tensor(data):
+            concat = lambda x,y,axis: torch.concatenate((x,y),dim=axis)
+            dt = torch
+        else:
+            concat = lambda x,y,axis: np.concatenate((x,y),axis=axis)
+            dt = np
+
+        idx = dt.zeros(data.idx_map.shape)
+        data.idx_map = concat(idx,data.idx_map,axis=-1) 
+        #idx_map will now have the index of the trajectory as well
+        for i in range(1,len(trajectories)):
+
+            new_traj = traj_to_contexts(trajectories[i],context_window_len=context_length,time_lag=time_lag)
+            data.data = concat(data.data,new_traj.data,axis=0)
+
+            #update index map for consistency
+            idx = np.zeros(new_traj.idx_map.shape) + i 
+            new_traj.idx_map = concat(idx,new_traj.idx_map,axis=-1)
+
+            data.idx_map = concat(data.idx_map,new_traj.idx_map,axis=0)
+
+        return data
 
 
 def traj_to_contexts(