Newer
Older
from pathlib import Path
import json
import numpy as np
import pandas as pd
import plotly.express as px
from typing import Union, Literal
def load_data(bench_suite):
for bench_file in Path(f"benchmarks/results").glob(f"*{bench_suite}*.json"):
(_, _, jvm, run) = bench_file.stem.split("-")
for data in json.loads(bench_file.read_bytes()):
benchmark = data["benchmark"].replace(
f"benchmarks.{bench_suite}.", "")
if "params" in data:
benchmark += "(" + ",".join(data["params"].values()) + ")"
for fork, times in enumerate(data["primaryMetric"]["rawData"]):
for time in times:
yield (jvm, benchmark, int(run), int(fork), time)
Mode = Union[Literal["bar"], Literal["strip"], Literal["box"]]
Orientation = Union[Literal["h"], Literal["v"]]
def make_bars_graph(bench_suite, max, mode: Mode, orientation: Orientation, scale: Scale):
df = pd.DataFrame(
load_data(bench_suite),
columns=("jvm", "benchmark", "run", "fork", "time")
)
if mode == "bar":
df = df.drop(columns=["run", "fork"])
df = df.groupby(["jvm", "benchmark"])
df = df.agg([("median", np.median), ("min", np.min), ("max", np.max)])
df = df.reset_index()
df.columns = [' '.join(col).strip() for col in df.columns.values]
df["time error minus"] = df["time median"] - df["time min"]
df["time error"] = df["time max"] - df["time median"]
reverse_order = orientation == "v"
df = df.sort_values(by=["jvm"], ascending=not reverse_order)
df = df.sort_values(by=["benchmark"], ascending=reverse_order)
args = {
"x" if orientation == "h" else "y": "time median" if mode == "bar" else "time",
"y" if orientation == "h" else "x": "benchmark",
"color": "jvm",
"orientation": orientation,
"color_discrete_sequence": px.colors.qualitative.Pastel1,
"template": "plotly_white"
}
if scale == "log":
args["log_x" if orientation == "h" else "log_y"] = True
elif scale == "lin":
args["range_x" if orientation == "h" else "range_y"] = [0, max]
if mode == "bar":
args |= {
"barmode": "group",
"error_x" if orientation == "h" else "error_y": "time error",
"error_x_minus" if orientation == "h" else "error_y_minus": "time error minus"
}
fig = getattr(px, mode)(df, **args)
log_note = "" if scale == "lin" else " Note the logarithmic scale."
"xaxis_title" if orientation == "h" else "yaxis_title": f"Throughput [ops/s] (Higher is better.{log_note})",
"yaxis_title" if orientation == "h" else "xaxis_title": "Benchmark",
"legend_traceorder": "reversed"
}
fig.update_layout(**layout_args)
fig.write_html(
f"benchmarks/graphs/{bench_suite}-{mode}-{orientation}-{scale}.html",
f"benchmarks/graphs/{bench_suite}-{mode}-{orientation}-{scale}.svg",
make_bars_graph("AppendBenchmark", 450000000, "bar", "h", "log")
make_bars_graph("AppendBenchmark", 450000000, "box", "h", "log")
make_bars_graph("AppendBenchmark", 450000000, "strip", "h", "log")
make_bars_graph("SumBenchmark", 450000, "bar", "h", "lin")
make_bars_graph("SumBenchmark", 450000, "box", "h", "lin")
make_bars_graph("SumBenchmark", 450000, "strip", "h", "lin")