Support tig for isolated nodes.

setup.py

@@ -5,7 +5,7 @@
 
 setup(
     name="sgw_tools",
-    version="1.104",
+    version="1.105",
     author="Mark Hale",
     license="MIT",
     description="Spectral graph wavelet tools",

sgw_tools/__init__.py

@@ -80,10 +80,14 @@ def _tig(g, s, **kwargs):
     if s.ndim == 1:
         s = s[..., np.newaxis]
     assert s.ndim == 2, "Signal shape was {}".format(s.shape)
+    if g.G.N == 1:
+        kwargs['method'] = 'exact'
     tig = g.filter(s[..., np.newaxis], **kwargs)
-    if s.shape[1] == 1: # single signal
+    if s.shape[1] == 1:  # single signal
         tig = tig[..., np.newaxis]
-    if tig.ndim == 2: # single filter
+    if tig.ndim == 1:
+        tig = tig[np.newaxis, ..., np.newaxis]
+    elif tig.ndim == 2:  # single filter
         tig = tig[..., np.newaxis]
     assert tig.shape == s.shape + (tig.shape[2],), "Tig shape was {}".format(tig.shape)
     return tig

tests/test_biggraph.py

@@ -200,20 +200,21 @@ def test_estimate_lmin(self):
         np.testing.assert_approx_equal(graph.lmin, 0.08818931)
 
         # isolated node
-        graph = BigGraph([[3]])
-        np.testing.assert_approx_equal(graph.lmin, np.nan)
+        for lap_type in ['combinatorial', 'normalized', 'adjacency']:
+            graph = BigGraph([[3]], lap_type=lap_type)
+            np.testing.assert_approx_equal(graph.lmin, np.nan)
 
     def test_estimate_lmax(self):
         graph = BigGraph.create_from(graphs.Sensor())
         self.assertRaises(ValueError, graph.estimate_lmax, method='unk')
 
         def check_lmax(graph, lmax):
             graph.estimate_lmax(method='bounds')
-            np.testing.assert_allclose(graph.lmax, lmax)
+            np.testing.assert_allclose(graph.lmax, lmax, err_msg='bounds', atol=1e-15)
             graph.estimate_lmax(method='lanczos')
-            np.testing.assert_allclose(graph.lmax, lmax)
+            np.testing.assert_allclose(graph.lmax, lmax, err_msg='lanczos', atol=1e-15)
             graph.compute_fourier_basis()
-            np.testing.assert_allclose(graph.lmax, lmax)
+            np.testing.assert_allclose(graph.lmax, lmax, err_msg='fourier', atol=1e-15)
 
         # Full graph (bound is tight).
         n_nodes, value = 10, 2
@@ -242,8 +243,9 @@ def check_lmax(graph, lmax):
         check_lmax(graph, lmax=2)
 
         # isolated node
-        graph = BigGraph([[3]])
-        check_lmax(graph, lmax=0)
+        for lap_type in ['combinatorial', 'normalized', 'adjacency']:
+            graph = BigGraph([[3]], lap_type=lap_type)
+            check_lmax(graph, lmax=0)
 
     def test_fourier_basis(self):
         # Smallest eigenvalue close to zero.

tests/test_sgw.py

@@ -0,0 +1,35 @@
+import unittest
+
+import numpy as np
+import sgw_tools as sgw
+import pygsp as gsp
+
+
+class TestCase(unittest.TestCase):
+    def test_tig(self):
+        G = sgw.BigGraph([[0,1],[1,0]])
+        g = gsp.filters.Heat(G)
+        s = sgw.createSignal(G)
+        sgw._tig(g, s)
+
+        # single signal
+        s = np.array([[1],[-1]])
+        sgw._tig(g, s)
+
+        # multiple signals
+        s = np.array([[1,2,3],[-1,-2,-3]])
+        sgw._tig(g, s)
+
+        # isolated node
+        G = sgw.BigGraph([[2]])
+        g = gsp.filters.Heat(G)
+        s = sgw.createSignal(G)
+        sgw._tig(g, s)
+
+        # single signal
+        s = np.array([1])
+        sgw._tig(g, s)
+
+        # multiple signals
+        s = np.array([[1,2,3]])
+        sgw._tig(g, s)