see if removing einsum fixes nan issues

rayflare/matrix_formalism/multiply_matrices.py

@@ -6,7 +6,7 @@
 # Contact: p.pearce@unsw.edu.au
 
 import numpy as np
-from sparse import load_npz, COO, stack, einsum
+from sparse import load_npz, COO, stack, einsum, dot
 from rayflare.angles import make_angle_vector, fold_phi, overall_bin
 import os
 import xarray as xr
@@ -162,17 +162,31 @@ def make_D(alphas, thick, thetas):
 # (GitHub: arsonwong)
 def dot_wl(mat, vec):
 
+    # if len(mat.shape) == 3:
+    #     result = einsum('ijk,ik->ij', mat, COO(vec)).todense()
+    #
+    # if len(mat.shape) == 2:
+    #     result = einsum('jk,ik->ij', mat, COO(vec)).todense()
+    result = np.empty((vec.shape[0], mat.shape[1]))
+
     if len(mat.shape) == 3:
-        result = einsum('ijk,ik->ij', mat, COO(vec)).todense()
+        for i1 in range(vec.shape[0]):  # loop over wavelengths
+            result[i1, :] = dot(mat[i1], vec[i1])
 
     if len(mat.shape) == 2:
-        result = einsum('jk,ik->ij', mat, COO(vec)).todense()
+        for i1 in range(vec.shape[0]):  # loop over wavelengths
+            result[i1, :] = dot(mat, vec[i1])
+
 
     return result
 
 
 def dot_wl_u2d(mat, vec):
-    result = einsum('jk,ik->ij', mat, COO(vec)).todense()
+    # result = einsum('jk,ik->ij', mat, COO(vec)).todense()
+    result = np.empty((vec.shape[0], vec.shape[1]))
+    for i1 in range(vec.shape[0]):  # loop over wavelengths
+        result[i1, :] = dot(mat, vec[i1])
+
     return result