diff --git a/pycs/astro/wl/mass_mapping.py b/pycs/astro/wl/mass_mapping.py
index e53f256..a879776 100755
--- a/pycs/astro/wl/mass_mapping.py
+++ b/pycs/astro/wl/mass_mapping.py
@@ -1200,6 +1200,11 @@ def sparse_wiener_filtering(
         )
 
         RMS_ShearMap = np.sqrt(InshearData.Ncov / 2.0)  # shape = (nx, ny)
+        # shape = ([nimgs], [Nrea], nx, ny)
+        # TODO: complex or complex128? Same question for real-valued arrays
+        xg = np.zeros_like(gamma1, dtype=complex)  # Gaussian + sparse components
+        xs = np.zeros_like(gamma1, dtype=complex)  # sparse component
+        xw = np.zeros_like(gamma1, dtype=complex)  # Gaussian component
         SigmaNoise = np.min(RMS_ShearMap)  # float
         Esn_Sparse = SigmaNoise / RMS_ShearMap  # shape = (nx, ny)
         Esn_Sparse[Esn_Sparse == np.inf] = 0
@@ -1250,11 +1255,7 @@ def sparse_wiener_filtering(
                 InshearData, mask, Nrea=Nrea, inpshape=inpshape
             )  # shape = ([nimgs], [Nrea], nx, ny)
 
-        # shape = ([nimgs], [Nrea], nx, ny)
-        # TODO: complex or complex128? Same question for real-valued arrays
-        xg = np.zeros_like(gamma1, dtype=complex)  # Gaussian + sparse components
-        xs = np.zeros_like(gamma1, dtype=complex)  # sparse component
-        xw = np.zeros_like(gamma1, dtype=complex)  # Gaussian component
+        
 
         for n in range(niter):
             resi1, resi2 = self.get_resi(
diff --git a/pycs/sparsity/sparse2d/starlet.py b/pycs/sparsity/sparse2d/starlet.py
index 352490d..38f342c 100644
--- a/pycs/sparsity/sparse2d/starlet.py
+++ b/pycs/sparsity/sparse2d/starlet.py
@@ -903,13 +903,13 @@ def threshold(
         if dim in (0, 2):  # SigmaNoise: scalar or shape = (nx, ny)
             # The noise level is obtained at each scale by multiplying by self.TabNorm
             Thres = (
-                SigmaNoise * (self.TabNsigma * self.TabNorm)[:, np.newdim, np.newdim]
+                SigmaNoise * (self.TabNsigma * self.TabNorm)[:, np.newaxis, np.newaxis]
             )  # shape = (ns, nx, ny)
         elif dim == 1:  # SigmaNoise: shape = (ns,)
             Thres = SigmaNoise * self.TabNsigma  # shape = (ns,)
         else:  # SigmaNoise: shape = (ns, nx, ny)
             Thres = (
-                SigmaNoise * self.TabNsigma[:, np.newdim, np.newdim]
+                SigmaNoise * self.TabNsigma[:, np.newaxis, np.newaxis]
             )  # shape = (ns, nx, ny)
 
         if ThresCoarse: