Added an 'identity' tensor splitting option

quimb/tensor/decomp.py

@@ -185,6 +185,18 @@
     return U, None, VH
 
 
+@compose
+def identity(x, **kwargs):
+    """
+    No-op "decomposition" that leaves the input unchanged. Can be useful to quickly build a tensor network representing a given tensor "as is".
+    """
+
+    if x.shape[0] < x.shape[1]:
+        return do("eye", x.shape[0], like=x), None, x
+    else:
+        return x, None, do("eye", x.shape[1], like=x)
+
+
 @compose
 def svd_truncated(
     x,

quimb/tensor/tensor_core.py

@@ -325,6 +325,7 @@ def rand_uuid(base=""):
 
 _VALID_SPLIT_GET = {None, "arrays", "tensors", "values"}
 _SPLIT_FNS = {
+    "identity": decomp.identity,
     "svd": decomp.svd_truncated,
     "eig": decomp.svd_via_eig_truncated,
     "lu": decomp.lu_truncated,