plot y axis as 'queries per second'

(instead of 'time per query' with inverted scale - marginally more logical)

plot.py

@@ -23,8 +23,8 @@
 for algo, color in zip(all_data.keys(), colors):
     data = all_data[algo]
     data.sort(key=lambda t: t[-2]) # sort by time
-    xs = [t[-2] for t in data]
-    ys = [t[-1] for t in data]
+    ys = [1.0 / t[-2] for t in data] # queries per second
+    xs = [t[-1] for t in data]
     ls = [t[0] for t in data]
     plt.plot(xs, ys, 'o', label=algo, color=color, markersize=3)
     #for i, l in enumerate(ls):
@@ -37,20 +37,20 @@
         y = t[-1]
         if y > last_y:
             last_y = y
-            ys.append(t[-1])
-            xs.append(t[-2])
+            xs.append(t[-1])
+            ys.append(1.0 / t[-2])
     handle, = plt.plot(xs, ys, 'o-', label=algo, color=color)
     handles.append(handle)
     labels.append(algo)
 
-plt.gca().set_xscale('log')
-plt.gca().set_title('Precision-Performance tradeoff - up and to the left is better')
-plt.gca().set_xlabel('Time per query (s) - lower is better')
-plt.gca().set_ylabel('10-NN precision - higher is better')
+plt.gca().set_yscale('log')
+plt.gca().set_title('Precision-Performance tradeoff - up and to the right is better')
+plt.gca().set_ylabel('Queries per second ($s^{-1}$) - larger is better')
+plt.gca().set_xlabel('10-NN precision - larger is better')
 box = plt.gca().get_position()
 plt.gca().set_position([box.x0, box.y0, box.width * 0.8, box.height])
 plt.gca().legend(handles, labels, loc='center left', bbox_to_anchor=(1, 0.5))
 plt.grid(b=True, which='both', color='0.65',linestyle='-')
-plt.ylim([0, 1.03])
+plt.xlim([0.0, 1.03])
 plt.savefig(args.output, bbox_inches='tight')
 

results/sift.txt

@@ -59,3 +59,4 @@ nearpy	NearPy(n_bits=16, hash_counts=40)	2689.25678992	1.5009472332	0.9558
 flann	FLANN(target_precision=0.500000)	909.874325991	0.000199952125549	0.2007
 bruteforce	BruteForce()	0.203589916229	0.744261413097	1.0
 flann	FLANN(target_precision=0.970000)	1713.85552406	0.00541697478294	0.8703
+nearpy	NearPy(n_bits=16, hash_counts=5)	420.568459034	0.211184137106	0.4588