Format units on plots (#156)

conbench/app/_plots.py

@@ -4,6 +4,7 @@
 import dateutil
 
 from ..hacks import sorted_data
+from ..units import formatter_for_unit
 
 
 class TimeSeriesPlotMixin:
@@ -30,9 +31,9 @@ def get_display_unit(unit):
     if unit == "s":
         return "seconds"
     elif unit == "B/s":
-        return "bytes/seconds"
+        return "bytes/second"
     elif unit == "i/s":
-        return "items/seconds"
+        return "items/second"
     else:
         return unit
 
@@ -62,82 +63,140 @@ def get_date_format():
     )
 
 
+def _simple_source(data, unit):
+    unit_fmt = formatter_for_unit(unit)
+    cases, points, means = [], [], []
+    points_formated, units_formatted = [], set()
+
+    for *values, point in data:
+        cases.append("-".join(values))
+        points.append(point)
+        formatted = unit_fmt(float(point), unit)
+        means.append(formatted)
+        formated_point, formatted_unit = formatted.split(" ", 1)
+        points_formated.append(formated_point)
+        units_formatted.add(formatted_unit)
+
+    unit_formatted = units_formatted.pop() if len(units_formatted) == 1 else None
+    if unit_formatted:
+        points = points_formated
+
+    axis_unit = unit_formatted if unit_formatted is not None else unit
+    if axis_unit == unit:
+        axis_unit = get_display_unit(unit)
+
+    source_data = dict(x=cases, y=points, means=means)
+    return bokeh.models.ColumnDataSource(data=source_data), axis_unit
+
+
 def simple_bar_plot(benchmarks, height=400, width=400):
     if len(benchmarks) > 30:
         return None
     if len(benchmarks) == 1:
         return None
 
     name = benchmarks[0]["tags"]["name"]
-    unit = get_display_unit(benchmarks[0]["stats"]["unit"])
-
-    cases, times = [], []
+    unit = benchmarks[0]["stats"]["unit"]
     data = sorted_data(benchmarks)
-    for *values, timing in data:
-        cases.append("-".join(values))
-        times.append(timing)
+    source, axis_unit = _simple_source(data, unit)
+
+    tooltips = [("mean", "@means")]
+    hover = bokeh.models.HoverTool(tooltips=tooltips)
 
     p = bokeh.plotting.figure(
-        x_range=cases,
+        x_range=source.data["x"],
         title=get_title(benchmarks, name),
         toolbar_location=None,
         plot_height=height,
         plot_width=width,
-        tools="",
+        tools=[hover],
+    )
+    p.vbar(
+        x="x",
+        top="y",
+        source=source,
+        width=0.9,
+        line_color="white",
+        color="silver",
     )
-    p.vbar(x=cases, top=times, width=0.9, line_color="white", color="silver")
+
     p.y_range.start = 0
     p.x_range.range_padding = 0.1
     p.xgrid.grid_line_color = None
     p.xaxis.major_label_orientation = 1
-    p.yaxis.axis_label = unit
+    p.yaxis.axis_label = axis_unit
 
     return p
 
 
-def time_series_plot(history, benchmark_id, height=250, width=1000):
-    unit = get_display_unit(history[0]["unit"])
-    current = [h for h in history if h["benchmark_id"] == benchmark_id]
-    with_dist = [h for h in history if h["distribution_mean"]]
-
-    times = [h["mean"] for h in history]
-    commits = [h["message"] for h in history]
-    dates = [dateutil.parser.isoparse(h["timestamp"]) for h in history]
-
-    times_x = [c["mean"] for c in current]
-    commits_x = [c["message"] for c in current]
-    dates_x = [dateutil.parser.isoparse(c["timestamp"]) for c in current]
-
-    times_mean = [w["distribution_mean"] for w in with_dist]
-    commits_mean = [w["message"] for w in with_dist]
-    dates_mean = [dateutil.parser.isoparse(w["timestamp"]) for w in with_dist]
-
-    alert_min, alert_max = [], []
-    for w in with_dist:
-        alert = 5 * float(w["distribution_stdev"])
-        alert_min.append(float(w["distribution_mean"]) - alert)
-        alert_max.append(float(w["distribution_mean"]) + alert)
+def _should_format(data, unit):
+    unit_fmt = formatter_for_unit(unit)
+
+    units_formatted = set()
+    for x in data:
+        units_formatted.add(unit_fmt(float(x["mean"]), unit).split(" ", 1)[1])
+
+    unit_formatted = units_formatted.pop() if len(units_formatted) == 1 else None
+
+    should = unit_formatted is not None
+    axis_unit = unit_formatted if unit_formatted is not None else unit
+    if axis_unit == unit:
+        axis_unit = get_display_unit(unit)
+
+    return should, axis_unit
+
+
+def _source(
+    data,
+    unit,
+    formatted=False,
+    distribution_mean=False,
+    alert_min=False,
+    alert_max=False,
+):
+    key = "distribution_mean" if distribution_mean else "mean"
+    unit_fmt = formatter_for_unit(unit)
+    commits = [x["message"] for x in data]
+    dates = [dateutil.parser.isoparse(x["timestamp"]) for x in data]
+
+    points, means = [], []
+    if alert_min:
+        for x in data:
+            alert = 5 * float(x["distribution_stdev"])
+            points.append(float(x["distribution_mean"]) - alert)
+            means.append(unit_fmt(points[-1], unit))
+    elif alert_max:
+        for x in data:
+            alert = 5 * float(x["distribution_stdev"])
+            points.append(float(x["distribution_mean"]) + alert)
+            means.append(unit_fmt(points[-1], unit))
+    else:
+        points = [x[key] for x in data]
+        means = [unit_fmt(float(x[key]), unit) for x in data]
 
-    source_data = dict(x=dates, y=times, commit=commits)
-    source = bokeh.models.ColumnDataSource(data=source_data)
+    if formatted:
+        points = [x.split(" ")[0] for x in means]
 
-    source_data_x = dict(x=dates_x, y=times_x, commit=commits_x)
-    source_x = bokeh.models.ColumnDataSource(data=source_data_x)
+    source_data = dict(x=dates, y=points, commits=commits, means=means)
+    return bokeh.models.ColumnDataSource(data=source_data)
 
-    source_data_mean = dict(x=dates_mean, y=times_mean, commit=commits_mean)
-    source_mean = bokeh.models.ColumnDataSource(data=source_data_mean)
 
-    source_data_alert_min = dict(x=dates_mean, y=alert_min, commit=commits_mean)
-    source_alert_min = bokeh.models.ColumnDataSource(data=source_data_alert_min)
+def time_series_plot(history, benchmark_id, height=250, width=1000):
+    unit = history[0]["unit"]
+    current = [h for h in history if h["benchmark_id"] == benchmark_id]
+    with_dist = [h for h in history if h["distribution_mean"]]
+    formatted, axis_unit = _should_format(history, unit)
 
-    source_data_alert_max = dict(x=dates_mean, y=alert_max, commit=commits_mean)
-    source_alert_max = bokeh.models.ColumnDataSource(data=source_data_alert_max)
+    source = _source(history, unit, formatted=formatted)
+    source_x = _source(current, unit, formatted=formatted)
+    source_mean = _source(with_dist, unit, formatted=formatted, distribution_mean=True)
+    source_alert_min = _source(with_dist, unit, formatted=formatted, alert_min=True)
+    source_alert_max = _source(with_dist, unit, formatted=formatted, alert_max=True)
 
     tooltips = [
         ("date", "$x{%F}"),
-        ("mean", "$y{0.000}"),
-        ("unit", unit),
-        ("commit", "@commit"),
+        ("mean", "@means"),
+        ("commit", "@commits"),
     ]
     hover = bokeh.models.HoverTool(
         tooltips=tooltips,
@@ -153,7 +212,7 @@ def time_series_plot(history, benchmark_id, height=250, width=1000):
 
     p.xaxis.formatter = get_date_format()
     p.xaxis.major_label_orientation = 1
-    p.yaxis.axis_label = unit
+    p.yaxis.axis_label = axis_unit
 
     p.line(source=source, legend_label="History")
     p.line(source=source_mean, color="#ffa600", legend_label="Mean")