use fftfreq for the frequencies

also set as property and remove ability to select value by call (can do
it by indexing now)

This affects the computation of Cdelta, which is now about 0.768 for the
Morlet, without having to set the values of s0 and dj lower. Setting
them lower doesn't change this much. See #1.

wavelets/transform.py

@@ -329,33 +329,16 @@ def compute_optimal_scales(self):
         sj = s0 * 2 ** (dj * np.arange(0, J + 1))
         return sj
 
-    # TODO: use np.frompyfunc on this
-    # TODO: can we just replace it with fftfreqs?
-    def w_k(self, k=None):
+    @property
+    def w_k(self):
         """Angular frequency as a function of fourier index.
 
         If no k, returns an array of all the angular frequencies
         calculated using the length of the data.
 
         See eq5 of TC.
         """
-        dt = self.dt
-        N = self.N
-        a = 2 * np.pi / (N * dt)
-        if k is None:
-            k = np.arange(N)
-            w_k = np.arange(N) * a
-            w_k[np.where(k > N // 2)] *= -1
-        elif type(k) is np.ndarray:
-            w_k = a * k
-            w_k[np.where(k > N // 2)] *= -1
-        else:
-            w_k = a * k
-            if k <= N // 2:
-                pass
-            elif k > N // 2:
-                w_k *= -1
-        return w_k
+        return np.fft.fftfreq(self.N, self.dt)
 
     @property
     def wavelet_transform(self):
@@ -507,7 +490,7 @@ def wavelet_transform_delta(self):
         """
         Y_0 = self.wavelet.frequency  # wavelet as f(w_k, s)
 
-        WK, S = np.meshgrid(self.w_k(), self.scales)
+        WK, S = np.meshgrid(self.w_k, self.scales)
 
         # compute Y_ over all s, w_k and sum over k
         norm = (2 * np.pi * S / self.dt) ** .5  # normalisation factor with dt=1