Add runtime graph.

2018/README.md

@@ -28,3 +28,11 @@ graph below. Can you guess which days had a large simulation?
 The code to produce this graph is included in the `other` directory, and
 should be compatible with most peoples repo set-ups, so you can share
 your pain. If it is not, please let me know.
+
+## Performance
+
+I benchmarked the performance of my implementations on stable rust
+(1.31.1 at the time of writing). The logarithmic runtime graph looks as
+follows:
+
+![Runtime by day](runtime.svg)

2018/other/runtimes.py

@@ -0,0 +1,52 @@
+#!/usr/bin/env python3
+import fileinput
+import re
+import numpy as np
+import matplotlib.pyplot as plt
+
+
+def parse_time(val):
+    val = val.replace(",", "")
+    return int(val) / 1e9
+
+
+def parse_input():
+    matcher = re.compile(r"day(\d{2})::part(1|2) ... bench:(\s+)([0-9,]+) ns/iter \(\+/- ([0-9,]+)\)$")
+    results = np.zeros((25, 2, 2))
+    for line in sorted(fileinput.input()):
+        match = matcher.search(line)
+        assert match
+
+        day = int(match.group(1)) - 1
+        part = int(match.group(2)) - 1
+        time = parse_time(match.group(4))
+        variance = parse_time(match.group(5))
+        results[day, part, :] = [time, variance]
+
+    return results
+
+
+def main():
+    results = parse_input()
+    _, ax = plt.subplots()
+    ax.set_yscale('log')
+    ax.set_xlabel('Day')
+    ax.set_ylabel('Runtime (s)')
+    ax.set_xticks(list(range(1, 26)))
+
+    bottom = np.min(results)
+
+    for i in range(results.shape[1]):
+        values = results[:, i, 0]
+        errors = results[:, i, 1]
+        pos = np.arange(len(values)) + 0.8 + 0.4 * i
+        ax.bar(pos, values, yerr=errors, align='center', width=0.4,
+               bottom=bottom)
+
+    ax.legend(['Part 1', 'Part 2'])
+
+    plt.savefig('runtime.svg')
+
+
+if __name__ == '__main__':
+    main()