backup

farfield_testing.py

@@ -8,6 +8,7 @@
 from autograd import grad, value_and_grad
 from autograd.test_util import check_grads
 from autograd.tracer import getval
+from scipy.ndimage import gaussian_filter
 
 import tidy3d as td
 from tidy3d.web import run
@@ -31,20 +32,20 @@ def get_spatial_coords_dict(simulation: td.Simulation, monitor: td.Monitor, fiel
     return coords
 
 
-def make_field_data(x, sim, monitor, rng) -> td.FieldData:
+def make_field_data_random(x, sim, monitor, rng) -> td.FieldData:
     field_cmps = {}
     grid = sim.discretize_monitor(monitor)
     for field_name in monitor.fields:
         coords = get_spatial_coords_dict(sim, monitor, field_name)
         coords["f"] = list(monitor.freqs)
 
-        data_shape = [len(coords[k]) for k in td.ScalarFieldDataArray._dims]
+        data_shape = np.array([len(coords[k]) for k in td.ScalarFieldDataArray._dims])
+        dims = np.where(data_shape > 1)[0]
 
-        x = x[:, None]
-        s = (1, np.prod(data_shape))
-        re = np.mean(x @ rng.uniform(-1, 1, s), axis=0)
-        im = np.mean(x @ rng.uniform(-1, 1, s), axis=0)
-        data = np.reshape(re + 1j * im, data_shape)
+        re = gaussian_filter(rng.uniform(-10, 10, min(data_shape[dims])), 1)
+        im = gaussian_filter(rng.uniform(-10, 10, min(data_shape[dims])), 1)
+        data = np.outer(x[: x.size // 2] ** 2, re) + 1j * np.outer(x[x.size // 2 :] ** 2, im)
+        data = np.reshape(data, data_shape)
 
         field_cmps[field_name] = td.ScalarFieldDataArray(data, coords=coords)
 
@@ -57,11 +58,42 @@ def make_field_data(x, sim, monitor, rng) -> td.FieldData:
     )
 
 
-def make_sim_data(x, sim, seed):
-    rng = np.random.default_rng(seed)
-    data = [
-        make_field_data(x, sim, mnt, rng) for mnt in sim.monitors if type(mnt) is td.FieldMonitor
-    ]
+def make_field_data_from_file(x, monitor, sim_data) -> td.FieldData:
+    field_cmps = {}
+    field_data = sim_data[monitor.name]
+    for c in field_data.field_components:
+        field = getattr(field_data, c)
+        f = field.values
+        data = x[0] * np.real(f) + 1j * x[1] * np.imag(f)
+        field_cmps[c] = td.ScalarFieldDataArray(data, coords=field.coords)
+
+    return td.FieldData(
+        monitor=field_data.monitor,
+        symmetry=field_data.symmetry,
+        symmetry_center=field_data.symmetry_center,
+        grid_expanded=field_data.grid_expanded,
+        **field_cmps,
+    )
+
+
+def make_sim_data(x, sim=None, seed=None, fp=None):
+    if fp is None and sim is not None:
+        rng = np.random.default_rng(seed)
+        data = [
+            make_field_data_random(x, sim, mnt, rng, fp)
+            for mnt in sim.monitors
+            if type(mnt) is td.FieldMonitor
+        ]
+    elif fp is not None:
+        sim_data = td.SimulationData.from_file(fp)
+        sim = sim_data.simulation
+        data = [
+            make_field_data_from_file(x, mnt, sim_data)
+            for mnt in sim.monitors
+            if type(mnt) is td.FieldMonitor
+        ]
+    else:
+        raise ValueError("yeah no")
     return td.SimulationData(simulation=sim, data=data)
 
 
@@ -143,7 +175,7 @@ def objective(x, *, is_2d=False, local=False, emulated=False, seed=None):
     fp = f'autograd_projection_dbg_{"2d" if is_2d else "3d"}.hdf5'
     if emulated:  # static sim, traced sim data
         sim = make_simulation(1.0, is_2d=is_2d)
-        sim_data = make_sim_data(x, sim, seed or 3467643734)
+        sim_data = make_sim_data(x, sim, fp="simulation_data.hdf5")
     elif Path(fp).is_file() and local:
         sim = make_simulation(x, is_2d=is_2d)
         sim_data = td.SimulationData.from_file(fp)
@@ -169,9 +201,9 @@ def objective(x, *, is_2d=False, local=False, emulated=False, seed=None):
 def main():
     seed = 46435543
     np.random.seed(seed)  # global seed for autograd
-    x = np.random.uniform(-1, 1, 10)
+    x = np.random.uniform(-10, 10, 2 * 71)
 
-    # objective(x, emulated=True, seed=seed)
+    # print(grad(objective)(x, emulated=True, seed=seed))
     check_grads(objective, modes=["rev"], order=1)(x, emulated=True, seed=seed)