HOSVD: Finish output and test coverage

pyttb/hosvd.py

@@ -1,10 +1,21 @@
+"""Higher Order SVD Implementation"""
+import warnings
+from typing import List, Optional
+
 import numpy as np
 import scipy
 
 import pyttb as ttb
 
 
-def hosvd(tensor, tol, verbosity=1, dimorder=None, sequential=True, ranks=None):
+def hosvd(
+    tensor,
+    tol: float,
+    verbosity: float = 1,
+    dimorder: Optional[List[int]] = None,
+    sequential: bool = True,
+    ranks: Optional[List[int]] = None,
+):
     """Compute sequentially-truncated higher-order SVD (Tucker).
 
     Computes a Tucker decomposition with relative error
@@ -19,6 +30,16 @@ def hosvd(tensor, tol, verbosity=1, dimorder=None, sequential=True, ranks=None):
     dimorder: Order to loop through dimensions
     sequential: Use sequentially-truncated version
     ranks: Specify ranks to consider rather than computing
+
+    Example
+    -------
+    >>> data = np.array([[29, 39.], [63., 85.]])
+    >>> tol = 1e-4
+    >>> disable_printing = -1
+    >>> tensorInstance = ttb.tensor().from_data(data)
+    >>> result = hosvd(tensorInstance, tol, verbosity=disable_printing)
+    >>> ((result.full() - tensorInstance).norm() / tensorInstance.norm()) < tol
+    True
     """
     # In tucker als this is N
     d = tensor.ndims
@@ -54,11 +75,11 @@ def hosvd(tensor, tol, verbosity=1, dimorder=None, sequential=True, ranks=None):
         print(
             f"||X||^2 = {normxsqr: g}\n"
             f"tol = {tol: g}\n"
-            f"eigenvalue sum threshold = tol^2 ||X||^2 / d = {eigsumthresh: g}\n"
+            f"eigenvalue sum threshold = tol^2 ||X||^2 / d = {eigsumthresh: g}"
         )
 
     # Main Loop
-    factor_matrices = [None] * d
+    factor_matrices = [np.empty(1)] * d
     # Copy input tensor, shrinks every step for sequential
     Y = ttb.tensor.from_tensor_type(tensor)
 
@@ -97,8 +118,21 @@ def hosvd(tensor, tol, verbosity=1, dimorder=None, sequential=True, ranks=None):
     if sequential:
         G = Y
     else:
-        G = Y.ttm(Y, factor_matrices, transpose=True)
+        G = Y.ttm(factor_matrices, transpose=True)
 
     result = ttb.ttensor.from_data(G, factor_matrices)
-    # TODO final printout
+
+    if verbosity > 0:
+        diffnormsqr = ((tensor - result.full()) ** 2).collapse()
+        relnorm = np.sqrt(diffnormsqr / normxsqr)
+        print(f" Size of core: {G.shape}")
+        if relnorm <= tol:
+            print(f"||X-T||/||X|| = {relnorm: g} <=" f"{tol: f} (tol)")
+        else:
+            print(
+                "Tolerance not satisfied!! "
+                f"||X-T||/||X|| = {relnorm: g} >="
+                f"{tol: f} (tol)"
+            )
+            warnings.warn("Specified tolerance was not achieved")
     return result

pyttb/tensor.py

@@ -1060,7 +1060,7 @@ def ttm(
             return Y
 
         if not isinstance(matrix, np.ndarray):
-            assert False, "matrix must be of type numpy.ndarray"
+            assert False, f"matrix must be of type numpy.ndarray but got:\n{matrix}"
 
         if not (dims.size == 1 and np.isin(dims, np.arange(self.ndims))):
             assert False, "dims must contain values in [0,self.dims)"

tests/test_hosvd.py

@@ -14,31 +14,49 @@ def sample_tensor():
 
 
 @pytest.mark.indevelopment
-def test_tucker_als_tensor_simple_convergence(capsys, sample_tensor):
+def test_hosvd_simple_convergence(capsys, sample_tensor):
     (data, T) = sample_tensor
     tol = 1e-4
     result = ttb.hosvd(T, tol)
-    assert (result.full()-T).norm()/T.norm() < tol, (
-        f"Failed to converge"
-    )
+    assert (result.full() - T).norm() / T.norm() < tol, f"Failed to converge"
+
+    tol = 1e-4
+    result = ttb.hosvd(T, tol, sequential=False)
+    assert (
+        result.full() - T
+    ).norm() / T.norm() < tol, f"Failed to converge for non-sequential option"
+
+    impossible_tol = 1e-20
+    with pytest.warns(UserWarning):
+        result = ttb.hosvd(T, impossible_tol)
+    assert (
+        result.full() - T
+    ).norm() / T.norm() > impossible_tol, f"Converged beyond provided precision"
 
 
 @pytest.mark.indevelopment
-def test_tucker_als_tensor_default_init(capsys, sample_tensor):
+def test_hosvd_default_init(capsys, sample_tensor):
     (data, T) = sample_tensor
     _ = ttb.hosvd(T, 1)
 
 
 @pytest.mark.indevelopment
-def test_tucker_als_tensor_incorrect_ranks(capsys, sample_tensor):
+def test_hosvd_smoke_test_verbosity(capsys, sample_tensor):
+    """For now just make sure verbosity calcs don't crash"""
+    (data, T) = sample_tensor
+    ttb.hosvd(T, 1, verbosity=10)
+
+
+@pytest.mark.indevelopment
+def test_hosvd_incorrect_ranks(capsys, sample_tensor):
     (data, T) = sample_tensor
     ranks = list(range(T.ndims - 1))
     with pytest.raises(ValueError):
         _ = ttb.hosvd(T, 1, ranks=ranks)
 
 
 @pytest.mark.indevelopment
-def test_tucker_als_tensor_incorrect_dimorder(capsys, sample_tensor):
+def test_hosvd_incorrect_dimorder(capsys, sample_tensor):
     (data, T) = sample_tensor
     dimorder = list(range(T.ndims - 1))
     with pytest.raises(ValueError):