Merge pull request #60 from kc-ml2/DEV/main

Dev/main

meent/on_numpy/emsolver/convolution_matrix.py

@@ -213,15 +213,15 @@ def to_conv_mat_discrete(ucell, fourier_order_x, fourier_order_y, device=None, t
         e_conv_all = np.zeros((ucell_pmt.shape[0], ff, ff)).astype(type_complex)
         o_e_conv_all = np.zeros((ucell_pmt.shape[0], ff, ff)).astype(type_complex)
         if improve_dft:
-            minimum_pattern_size = 2 * ff * ucell_pmt.shape[2]
+            minimum_pattern_size = 2 * ff * ucell_pmt.shape[2]  # TODO: scale factor is 2? to avoid alias?
         else:
-            minimum_pattern_size = 2 * ff
+            minimum_pattern_size = 4 * fourier_order_x + 1  # TODO: other bds
 
         for i, layer in enumerate(ucell_pmt):
             n = minimum_pattern_size // layer.shape[1]
             layer = np.repeat(layer, n + 1, axis=1)
-            f_coeffs = np.fft.fftshift(np.fft.fft(layer / layer.size))
-            o_f_coeffs = np.fft.fftshift(np.fft.fft(1/layer / layer.size))
+            f_coeffs = np.fft.fftshift(np.fft.fft(layer / layer.size).astype(type_complex))
+            o_f_coeffs = np.fft.fftshift(np.fft.fft(1/layer / layer.size).astype(type_complex))
             # FFT scaling:
             # https://kr.mathworks.com/matlabcentral/answers/15770-scaling-the-fft-and-the-ifft?s_tid=srchtitle
 
@@ -257,8 +257,8 @@ def to_conv_mat_discrete(ucell, fourier_order_x, fourier_order_y, device=None, t
                 n = minimum_pattern_size_x // layer.shape[1]
                 layer = np.repeat(layer, n + 1, axis=1)
 
-            f_coeffs = np.fft.fftshift(np.fft.fft2(layer / layer.size))
-            o_f_coeffs = np.fft.fftshift(np.fft.fft2(1/layer / layer.size))
+            f_coeffs = np.fft.fftshift(np.fft.fft2(layer / layer.size).astype(type_complex))
+            o_f_coeffs = np.fft.fftshift(np.fft.fft2(1/layer / layer.size).astype(type_complex))
             center = np.array(f_coeffs.shape) // 2
 
             conv_idx_y = np.arange(-ff_y + 1, ff_y, 1)

meent/on_torch/emsolver/convolution_matrix.py

@@ -364,8 +364,8 @@ def to_conv_mat_raster_discrete(ucell, fourier_order_x, fourier_order_y, device=
         for i, layer in enumerate(ucell_pmt):
             n = minimum_pattern_size // layer.shape[1]
             layer = layer.repeat_interleave(n + 1, axis=1)
-            f_coeffs = torch.fft.fftshift(torch.fft.fft(layer / layer.numel()))
-            o_f_coeffs = torch.fft.fftshift(torch.fft.fft(1/layer / layer.numel()))
+            f_coeffs = torch.fft.fftshift(torch.fft.fft(layer / layer.numel()).type(type_complex))
+            o_f_coeffs = torch.fft.fftshift(torch.fft.fft(1/layer / layer.numel()).type(type_complex))
             center = torch.tensor(f_coeffs.shape, device=device) // 2
             # center = torch.div(torch.tensor(f_coeffs.shape, device=device), 2, rounding_mode='trunc')
             # center = torch.tensor(center, device=device)

meent/on_torch/emsolver/transfer_method.py

@@ -422,7 +422,7 @@ def transfer_2d_3(center, big_F, big_G, big_T, Z_I, Y_I, psi, theta, ff_xy, delt
 
     final_RT = torch.linalg.inv(final_A) @ final_B
 
-    R_s = final_RT[:ff_xy, :].flatten()
+    R_s = final_RT[:ff_xy, :].flatten()  # TODO: why flatten?
     R_p = final_RT[ff_xy:2 * ff_xy, :].flatten()
 
     big_T1 = final_RT[2 * ff_xy:, :]

meent/on_torch/mee.py

@@ -8,7 +8,7 @@
 
 class MeeTorch(ModelingTorch, RCWATorch, OptimizerTorch):
 
-    def __init__(self, device=None, type_complex=None, *args, **kwargs):
+    def __init__(self, device=0, type_complex=0, *args, **kwargs):
 
         # device
         if device in (0, 'cpu'):
@@ -35,6 +35,6 @@ def __init__(self, device=None, type_complex=None, *args, **kwargs):
         self.device = device
         self.type_complex = type_complex
 
-        ModelingTorch.__init__(self, *args, **kwargs)
-        RCWATorch.__init__(self, *args, **kwargs)
-        OptimizerTorch.__init__(self, *args, **kwargs)
+        ModelingTorch.__init__(self, device=device, type_complex=type_complex, *args, **kwargs)
+        RCWATorch.__init__(self, device=device, type_complex=type_complex, *args, **kwargs)
+        OptimizerTorch.__init__(self, device=device, type_complex=type_complex, *args, **kwargs)

setup.py

@@ -12,7 +12,7 @@
 }
 setup(
     name='meent',
-    version='0.9.6',
+    version='0.9.7',
     url='https://github.com/kc-ml2/meent',
     author='KC ML2',
     author_email='yongha@kc-ml2.com',