Add benchmarking plots

2021/README.md

@@ -20,3 +20,15 @@ OPTIONS:
     -i, --input <INPUT>    Read input from the given file instead of stdin
     -t, --time             Print time taken
 ```
+
+## That goal was achieved
+
+Runtime benchmarked with [Criterion], reading input directly from memory to avoid disk IO
+inconsistencies.
+
+![Cumulative time](./cumulative-time.svg)
+
+![Time by day](./individual-time.svg)
+
+
+[Criterion]: https://github.com/bheisler/criterion.rs

2021/create_timing_plots.py

@@ -0,0 +1,97 @@
+#!/usr/bin/env python3
+import json
+from pathlib import Path
+from typing import Dict
+
+import numpy as np
+import matplotlib.pyplot as plt
+
+
+def read_timings() -> Dict[int, Dict]:
+    timings = {}
+
+    for day in Path('target/criterion/part1').iterdir():
+        with open(day / 'new' / 'estimates.json', mode='rb') as f:
+            timings[int(day.parts[-1])] = {
+                1: json.load(f)
+            }
+
+    for day in Path('target/criterion/part2').iterdir():
+        with open(day / 'new' / 'estimates.json', mode='rb') as f:
+            timings[int(day.parts[-1])][2] = json.load(f)
+
+    return timings
+
+
+def plot_cumulative_time(timings: Dict[int, Dict]):
+    plt.clf()
+
+    times = [0]
+
+    for day in range(min(timings.keys()), max(timings.keys()) + 1):
+        times.append(timings[day][1]['mean']['point_estimate'])
+        if day < 25:
+            times.append(timings[day][2]['mean']['point_estimate'])
+        else:
+            times.append(0)
+
+    cumulative = np.cumsum(times)
+    # Convert from nanoseconds to seconds
+    cumulative /= 1e9
+
+    x = np.arange(0.0, 25.5, 0.5)
+
+    plt.plot(x, cumulative, label="Cumulative time", drawstyle='steps-post')
+    plt.plot([0, 25], [0, 0.5], label="Target time")
+    plt.ylabel('Cumulative time (s)')
+    plt.xlabel('Days completed')
+
+    plt.legend()
+    plt.tight_layout()
+
+    plt.xlim(0, 25)
+    plt.ylim(0, 0.5)
+
+    plt.savefig('cumulative-time.svg')
+
+
+def plot_individual_times(timings: Dict[int, Dict]):
+    plt.clf()
+
+    def plot(parts, **kwargs):
+        x = np.arange(1, len(parts) + 1)
+
+        values = np.array(list(part['mean']['point_estimate'] for part in parts))
+        upper = np.array(list(part['mean']['confidence_interval']['upper_bound'] for part in parts))
+        lower = np.array(list(part['mean']['confidence_interval']['lower_bound'] for part in parts))
+
+        # Convert from ns to s
+        yerr = np.array([upper - values, lower - values]) / 1e9
+        values = values / 1e9
+
+        plt.bar(x, values, yerr=yerr, align='edge', log=True, **kwargs)
+        pass
+
+    plot(list(timings[day][1] for day in range(1, 26)), label="Part 1", width=-0.4)
+    plot(list(timings[day][2] for day in range(1, 25)), label="Part 2", width=0.4)
+
+    plt.ylabel('Runtime (s)')
+    plt.xlabel('Day')
+
+    plt.xlim(0, 26)
+    plt.xticks(np.arange(1, 26))
+
+    plt.legend()
+    plt.tight_layout()
+
+    plt.savefig('individual-time.svg')
+
+
+def main():
+    timings = read_timings()
+    plot_cumulative_time(timings)
+    plot_individual_times(timings)
+
+
+if __name__ == '__main__':
+    main()