update expwin to be band-pass

doc/history.rst

@@ -22,6 +22,7 @@ Additional features:
 * print(graph) and print(filters) now show valuable information.
 * Building a graph object is much faster.
 * New rectangular filter (low-pass and band-pass).
+* The exponential window has been updated from low-pass only to band-pass.
 
 0.5.1 (2017-12-15)
 ------------------

pygsp/filters/__init__.py

@@ -61,6 +61,7 @@
 
 .. autosummary::
 
+    Expwin
     Rectangular
 
 Approximations

pygsp/filters/expwin.py

@@ -13,10 +13,14 @@ class Expwin(Filter):
     Parameters
     ----------
     G : graph
+    band_min : float
+        Minimum relative band. The filter take the value 0.5 at this relative
+        frequency (default = None).
     band_max : float
-        Maximum relative band (default = 0.2)
+        Maximum relative band. The filter take the value 0.5 at this relative
+        frequency (default = 0.2).
     slope : float
-        Slope parameter (default = 1)
+        The slope at cut-off (default = 1).
 
     Examples
     --------
@@ -27,36 +31,49 @@ class Expwin(Filter):
     >>> G = graphs.Ring(N=20)
     >>> G.estimate_lmax()
     >>> G.set_coordinates('line1D')
-    >>> g = filters.Expwin(G)
+    >>> g = filters.Expwin(G, band_min=0.1, band_max=0.5)
     >>> s = g.localize(G.N // 2)
     >>> fig, axes = plt.subplots(1, 2)
     >>> g.plot(ax=axes[0])
     >>> G.plot_signal(s, ax=axes[1])
 
     """
 
-    def __init__(self, G, band_max=0.2, slope=1):
+    def __init__(self, G, band_min=None, band_max=0.2, slope=1):
 
         self.band_max = band_max
+        self.band_min = band_min
         self.slope = slope
 
-        def fx(x):
+        def f(x):
             # Canary to avoid division by zero and overflow.
             y = np.where(x <= 0, -1, x)
             y = np.exp(-slope / y)
             return np.where(x <= 0, 0, y)
 
-        def gx(x):
-            y = fx(x)
-            return y / (y + fx(1 - x))
+        def h(x):
+            y = f(x)
+            z = f(1 - x)
+            return y / (y + z)
 
-        def ffin(x):
-            return gx(1 - x)
+        def g_highpass(x):
+            return h(0.5 - x/G.lmax + band_max)
+        def g_lowpass(x):
+            return h(0.5 + x/G.lmax - band_min)
 
-        g = [lambda x: ffin(x/band_max/G.lmax)]
+        if band_min is None:
+            g = [g_highpass]
+        elif band_max is None:
+            g = [g_lowpass]
+        else:
+            g = [lambda x: g_lowpass(x) * g_highpass(x)]
 
         super(Expwin, self).__init__(G, g)
 
     def _get_extra_repr(self):
-        return dict(band_max='{:.2f}'.format(self.band_max),
-                    slope='{:.2f}'.format(self.slope))
+        attrs = dict(slope='{:.0f}'.format(self.slope))
+        if self.band_min is not None:
+            attrs.update(band_min='{:.2f}'.format(self.band_min))
+        if self.band_max is not None:
+            attrs.update(band_max='{:.2f}'.format(self.band_max))
+        return attrs

pygsp/tests/test_filters.py

@@ -214,6 +214,12 @@ def test_heat(self):
     def test_expwin(self):
         f = filters.Expwin(self._G)
         self._test_methods(f, tight=False)
+        f = filters.Expwin(self._G, band_min=None, band_max=0.8)
+        self._test_methods(f, tight=False)
+        f = filters.Expwin(self._G, band_min=0.1, band_max=None)
+        self._test_methods(f, tight=False)
+        f = filters.Expwin(self._G, band_min=0.1, band_max=0.7)
+        self._test_methods(f, tight=False)
 
     def test_rectangular(self):
         f = filters.Rectangular(self._G)