Add Rhythm model

Author: Kominaru

collections/yes/MWC18Finals.txt

@@ -6,5 +6,5 @@ class Config():
     MODE_COLLECTIONS = 1
 
     plots = True        # Whether to plot stuff 
-    wcsv  = True         # Whether to write results into csv
-    mode  = MODE_RANKED
+    wcsv  = False        # Whether to write results into csv
+    mode  = MODE_COLLECTIONS

collections/yes/MWC21Finals.txt

@@ -0,0 +1,74 @@
+import numpy as np
+import math
+
+###############################################################################
+# Obtains the rhythm calculation for each note in the map by counting the
+# amount of notes around the note in a timing window of size bin_size
+###############################################################################
+bin_size = 1000
+
+# TO-DO: Make the window non-rigid by weighting over a bell curve over (-2,2) instead
+
+
+def gaussian(x):
+    return math.exp(-(x**2)/(2))
+
+
+def bind_500(x): return max(min(x, 500), -500)
+
+
+def chord_suppression(x):
+    grace_threshold = 30
+    slope = +0.2
+    x = bind_500(x)
+    return 1 / (1 + math.exp(grace_threshold*slope-slope*x))
+
+
+def weighted_avg_and_std(values, weights):
+    """
+    Return the weighted average and standard deviation.
+    values, weights -- Numpy ndarrays with the same shape.
+    """
+    average = np.average(values, weights=weights)
+    # Fast and numerically precise:
+    variance = np.average((values-average)**2, weights=weights)
+    return (average, math.sqrt(variance))
+
+
+def obtainRhythmCalculation(ho):
+    rhythm = np.ones(len(ho))
+    wl = 0  # Current index of the note that first enters in the window
+    wr = 0  # Same but for last
+
+    for i in range(len(ho)):
+        r = 0
+        # Find the new first note that is inside the window
+        while ho[wl].timestamp < (ho[i].timestamp-bin_size/2):
+            wl += 1
+
+        # Fin the new last note that is inside the window
+        while ho[wr].timestamp < (ho[i].timestamp+bin_size/2) and wr < len(ho)-1:
+            wr += 1
+
+        if wr-wl > 0:
+
+            # Raw distances between consecutive notes
+            distances = np.zeros(wr-wl)
+            # Gaussian weights assigned depending on distance to i, the center of the window (basically to smooth the calculations like in Density and Manip models)
+            weights = np.zeros(wr-wl)
+
+            # We always use the closest note to i as the note to calculate the weight with (gaussian(((ho[j+1]... vs. gaussian(((ho[j]...)
+            for j in range(wl, i):
+                distances[j-wl] = ho[j+1].timestamp-ho[j].timestamp
+                weights[j-wl] = gaussian(((ho[j+1].timestamp-ho[i].timestamp)/(
+                    bin_size/2))*3)*chord_suppression(ho[j+1].timestamp-ho[j].timestamp)
+            for j in range(i, wr):
+                distances[j-wl] = ho[j+1].timestamp-ho[j].timestamp
+                weights[j-wl] = gaussian(((ho[j].timestamp-ho[i].timestamp)/(
+                    bin_size/2))*3)*chord_suppression(ho[j+1].timestamp-ho[j].timestamp)
+
+            # Compute the weighted avg and std deviation, rhythm difficulty for that note is stdev/avg. Perhaps this metric is not correct and favors higher BPM /lower BPM/faster maps?
+            norm_avg, norm_stdev = weighted_avg_and_std(distances, weights)
+            rhythm[i] += norm_stdev/(1+norm_avg)
+
+    return rhythm