Merge main into working branch for PR.

.gitattributes

@@ -0,0 +1 @@
+*.msh filter=lfs diff=lfs merge=lfs -text

README.md

@@ -47,6 +47,23 @@ need to _separately_ ensure the solver backend is installed (again, currently al
 Alternatively (and this is important for large-scale RL training), the core Hydrogym package can (or will soon be able to) launch reinforcement learning training on a Ray-cluster without an underlying Firedrake install.
 For more information and suggested approaches see the [Installation Docs](https://hydrogym.readthedocs.io/en/latest/installation.html).
 
+To add HydroGym to an existing Firedrake installation, and install from the repository, run:
+
+```bash
+git clone https://github.com/dynamicslab/hydrogym.git
+cd hydrogym
+pip install .
+```
+
+As the mesh files are stored in [git large file storage](https://git-lfs.github.com/), you will need to install git-lfs
+to download the mesh files.
+
+```bash
+git lfs install && git lfs fetch --all
+```
+
+At which point you are ready to run HydroGym locally.
+
 # Quickstart Guide
 
  Having installed Hydrogym into our virtual environment experimenting with Hydrogym is as easy as starting the Python interpreter

examples/cylinder/pressure-probes.py

@@ -20,15 +20,8 @@
 mesh = "medium"
 method = "BDF"
 
-# Configure pressure probes - evenly spaced around the cylinder
-# n_probes = 8
-# R = 0.5
-# probes = [(R * np.cos(theta), R * np.sin(theta))
-#           for theta in np.linspace(0, 2 * np.pi, n_probes, endpoint=False)]
-
-observation_type = "velocity_probes"
-
 # Velocity probes
+observation_type = "velocity_probes"
 xp = np.linspace(1.0, 10.0, 16)
 yp = np.linspace(-2.0, 2.0, 4)
 X, Y = np.meshgrid(xp, yp)
@@ -39,7 +32,7 @@
     Re=100,
     mesh=mesh,
     velocity_order=velocity_order,
-    observation_type=observation_type,
+    observation_type="vorticity_probes",
     probes=probes,
 )
 
@@ -82,7 +75,7 @@ def controller(t, y):
     callbacks=callbacks,
     method=method,
     stabilization=stabilization,
-    # controller=controller,
+    # controller=controller, --> TODO: This needs to be done in a more principled fashion
 )
 
 data = np.loadtxt(data_file)

hydrogym/core.py

@@ -40,8 +40,7 @@ class PDEBase(metaclass=abc.ABCMeta):
   DEFAULT_MESH = ""
   DEFAULT_DT = np.inf
 
-  # Timescale used to smooth inputs
-  #  (should be less than any meaningful timescale of the system)
+  # Timescale used to smooth inputs (should be less than any meaningful timescale of the system)
   TAU = 0.0
 
   StateType = TypeVar("StateType")
@@ -359,7 +358,7 @@ def step(self, iter: int, control: Iterable[float] = None, **kwargs):
         """
     raise NotImplementedError
 
-  def reset(self, t=0.0):
+  def reset(self):
     """Reset variables for the timestepper"""
     self.t = 0.0
 
@@ -497,16 +496,8 @@ def check_complete(self):
 
   def reset(self, t=0.0) -> Union[ArrayLike, Tuple[ArrayLike, dict]]:
     self.iter = 0
-    self.t = 0.
-
-    if self.restart_ckpts is not None:
-      ckpt_index = np.random.randint(0, len(self.restart_ckpts))
-      self.flow.load_checkpoint(self.restart_ckpts[ckpt_index])
-      # print("Loaded ckeckpoint:", self.restart_ckpts[ckpt_index], flush=True)
-
-    self.flow.reset(q0=self.flow.copy_state() if self.restart_ckpts is not None else self.q0)
+    self.flow.reset(q0=self.q0, t=t)
     self.solver.reset()
-    info = {}
 
     return self.flow.get_observations(), info