tonybaloney · April 2, 2018 12:18 · Apr 2, 2018
diff --git a/plot_results.py b/plot_results.py
@@ -0,0 +1,54 @@
+import argparse
+from pathlib import Path
+from perf._bench import BenchmarkSuite
+
+import seaborn as sns
+import pandas as pd
+
+sns.set(style="whitegrid")
+
+parser = argparse.ArgumentParser(description='Convert a list of benchmarks into a CSV')
+parser.add_argument('files', metavar='N', type=str, nargs='+',
+                    help='files to compare')
+args = parser.parse_args()
+
+benchmark_names = []
+records = []
+first = True
+for f in args.files:
+    benchmark_suite = BenchmarkSuite.load(f)
+    if first:
+        # Initialise the dictionary keys to the benchmark names
+        benchmark_names = benchmark_suite.get_benchmark_names()
+        first = False
+    bench_name = Path(benchmark_suite.filename).name
+    for name in benchmark_names:
+        try:
+            benchmark = benchmark_suite.get_benchmark(name)
+            if benchmark is not None:
+                records.append({
+                    'test': name,
+                    'runtime': bench_name.replace('.json', ''),
+                    'stdev': benchmark.stdev(),
+                    'mean': benchmark.mean(),
+                    'median': benchmark.median()
+                })
+        except KeyError:
+            # Bonus benchmark! ignore.
+            pass
+
+df = pd.DataFrame(records)
+
+for test in benchmark_names:
+    # Draw a pointplot to show pulse as a function of three categorical factors
+    g = sns.factorplot(
+        x="runtime",
+        y="mean",
+        data=df[df['test'] == test],
+        #capsize=.2,
+        palette="YlGnBu_d",
+        size=12,
+        aspect=1,
+        kind="bar")
+    g.despine(left=True)
+    g.savefig("png/{}-result.png".format(test))