bug fix in masking experimental data.. tensorflow-nufft is now default.

mjo22 · Jul 31, 2023 · 921fb25 · 921fb25
1 parent cd33585
commit 921fb25
Show file tree

Hide file tree

Showing 3 changed files with 28 additions and 28 deletions.
diff --git a/src/jax_2dtm/simulator/image.py b/src/jax_2dtm/simulator/image.py
@@ -64,7 +64,7 @@ class Image(metaclass=ABCMeta):
     filters: InitVar[list[Filter] | None] = None
     masks: InitVar[list[Filter] | None] = None
     observed: InitVar[Array | None] = None
-    _process_observed: bool = True
+    _process_observed: bool = field(pytree_node=False, default=True)
 
     def __post_init__(self, filters, masks, observed):
         # Set filters
@@ -227,7 +227,7 @@ def sample(self, state: Optional[ParameterState] = None) -> Array:
         state = state or self.state
         simulated = self.render(state)
         noise = state.noise.sample(self.config.freqs * self.config.pixel_size)
-        return simulated + self.mask(noise)
+        return simulated + fft(self.mask(ifft(noise)))
 
     def log_likelihood(self, state: Optional[ParameterState] = None) -> Scalar:
         """Evaluate the log-likelihood of the data given a parameter set."""

diff --git a/src/jax_2dtm/simulator/optics.py b/src/jax_2dtm/simulator/optics.py
@@ -78,8 +78,8 @@ class CTFOptics(Optics):
     b_factor : `jax_2dtm.types.Scalar`
     """
 
-    defocus_u: Scalar = 8000.0
-    defocus_v: Scalar = 8000.0
+    defocus_u: Scalar = 10000.0
+    defocus_v: Scalar = 10000.0
     defocus_angle: Scalar = 0.0
     voltage: Scalar = 300.0
     spherical_aberration: Scalar = 2.7

diff --git a/src/jax_2dtm/utils/integration.py b/src/jax_2dtm/utils/integration.py
@@ -6,10 +6,10 @@
 
 import jax
 import jax.numpy as jnp
-#import tensorflow_nufft as tfft
-#from jax.experimental import jax2tf
+import tensorflow_nufft as tfft
+from jax.experimental import jax2tf
 
-from jax_finufft import nufft1
+# from jax_finufft import nufft1
 from jax.scipy import special
 
 from .fft import fftfreqs1d
@@ -57,15 +57,15 @@ def nufft(
     """
     complex_density = density.astype(complex)
     periodic_coords = 2 * jnp.pi * coords / box_size
-    #nufft1 = jax2tf.call_tf(
-    #    _tf_nufft1,
-    #    output_shape_dtype=jax.ShapeDtypeStruct(shape, complex_density.dtype),
-    #)
-    #ft = nufft1(
-    #    complex_density, jnp.flip(periodic_coords, axis=-1), shape, eps
-    #)
-    x, y = periodic_coords.T
-    ft = nufft1(shape, complex_density, -y, -x, eps=eps)
+    nufft1 = jax2tf.call_tf(
+        _tf_nufft1,
+        output_shape_dtype=jax.ShapeDtypeStruct(shape, complex_density.dtype),
+    )
+    ft = nufft1(
+        complex_density, jnp.flip(periodic_coords, axis=-1), shape, eps
+    )
+    # x, y = periodic_coords.T
+    # ft = nufft1(shape, complex_density, -y, -x, eps=eps)
 
     return ft
 
@@ -109,18 +109,18 @@ def integrate_gaussians(
     return image
 
 
-#def _tf_nufft1(source, points, shape, tol):
-#    """
-#    Wrapper for type-1 non-uniform FFT
-#    from tensorflow-nufft.
-#    """
-#    return tfft.nufft(
-#        source,
-#        points,
-#        grid_shape=shape,
-#        transform_type="type_1",
-#        tol=tol.numpy(),
-#    )
+def _tf_nufft1(source, points, shape, tol):
+    """
+    Wrapper for type-1 non-uniform FFT
+    from tensorflow-nufft.
+    """
+    return tfft.nufft(
+        source,
+        points,
+        grid_shape=shape,
+        transform_type="type_1",
+        tol=tol.numpy(),
+    )
 
 
 @jax.jit