generalised normalisation for autocrosscorr

time_series_analysis.py

@@ -5,11 +5,8 @@ import matplotlib.pylab as plt
 def autocrosscorr(
     yA,
     yB=None,
-    connected=False,
+    connected=True,
     normalised=True,
-    t=None,
-    figs=False,
-    plotmean=True,
 ):
     """
     Calculates normalised auto- or cross-correlations as a function of time.
@@ -28,13 +25,15 @@ def autocrosscorr(
     connected: boolean
         If True, find the connected correlation function, which measures the
         correlation once the population mean has been substracted.
-    normalise: boolean
-        If True, normalise each time point by the standard deviation across
-        the replicates
+    normalised: boolean
+        If True and connected is True, normalise each time point by the
+        standard deviation across the replicates.
+        If True and connected is False, normalise each time by the root mean
+        square across replicates.
 
     Returns
     -------
-    r_corr: array
+    corr: array
         An array of the correlations with each row the result for the
         corresponding replicate and each column a time point
     """
@@ -58,6 +57,7 @@ def autocrosscorr(
         stdB = np.sqrt(np.nanmean(dyB**2, axis=0).reshape((1, n)))
     else:
         # auto correlation
+        yB = yA
         dyB = dyA
         stdB = stdA
     # calculate correlation
@@ -67,11 +67,14 @@ def autocrosscorr(
         prods = [dyA[:, t] * dyB[:, t + r] for t in range(n - r)]
         corr[:, r] = np.nansum(prods, axis=0) / (n - r)
     if normalised:
-        r_corr = np.array(corr) / stdA / stdB
-    else:
-        r_corr = np.array(corr)
-    # return as an array
-    return r_corr
+        if connected:
+            # normalise by standard deviation
+            corr = corr / stdA / stdB
+        else:
+            # normalise by root mean square
+            corr = corr / np.sqrt(np.nanmean(yA**2, axis=1).reshape((nr, 1))
+                                  * np.nanmean(yB**2, axis=1).reshape((nr, 1)))
+    return corr
 
 
 ###