def percentiles_by_stencil(results, percentiles):
"""Computes given percentiles of measured run times of multiple results
and sorts them by stencil.
Args:
results: List of `Result` objects.
percentils: List of percentiles, each percentile must be in the range
[0, 100] (inclusive).
Returns:
A tuple of lists (stencils, percentiles 1, percentiles 2, ...). The
number of returned values depends on the number of input percentiles.
"""
stencils = results[0].stencils
if any(stencils != r.stencils for r in results):
raise ArgumentError('All results must include the same stencils')
qtimes = []
for q in percentiles:
def compute_q(times):
return np.percentile(times, q)
qtimes.append(by_stencils(r.mapped_times(compute_q) for r in results))
return (stencils, *qtimes)
def get_datetime(result):
if key == 'runtime':
datetime = result.runtime.datetime
elif key == 'job':
datetime = result.datetime
else:
raise ArgumentError('"key" argument must be "runtime" or "job"')
return time.local_time(datetime)
def times_by_stencil(results):
"""Collects times of multiple results by stencils.
Args:
results: List of `Result` objects.
Returns:
A tuple of lists (stencils, times).
"""
stencils = results[0].stencils
if any(stencils != r.stencils for r in results):
raise ArgumentError('All results must include the same stencils')
times = by_stencils(r.times_by_stencil() for r in results)
return stencils, times
def from_timestr(timestr):
"""Converts a time string back to a `datetime.datetime` object.
The given string must be in the format as generated by the `timestr`
function.
Args:
timestr: A string representing a time in the supported format.
Returns:
A `datetime.datetime` object, representing the same time as `timestr`.
"""
try:
return datetime.strptime(timestr, '%Y-%m-%dT%H:%M:%S.%f%z')
except ValueError:
raise ArgumentError(f'"{timestr}" is an invalid time string') from None
def statistics_by_stencil(results):
"""Computes mean and stdev times of multiple results by stencil.
Args:
results: List of `Result` objects.
Returns:
A tuple of lists (stencils, meantimes, stdevtimes).
"""
stencils = results[0].stencils
if any(stencils != r.stencils for r in results):
raise ArgumentError('All results must include the same stencils')
meantimes = by_stencils(r.mapped_times(np.mean) for r in results)
stdevtimes = by_stencils(r.mapped_times(np.std) for r in results)
return stencils, meantimes, stdevtimes
def timestr(time):
"""Returns the given time as a string.
Args:
time: The time that should be converted to a string.
Returns:
A string representing `time`, compatible to ISO 8601 time format.
"""
try:
dt_format = ('%Y-%m-%dT%H:%M:%S.%f%z', '%a %b %d %H:%M:&S %Z %Y', '%Y-%m-%d %H;%M')
for current_format in dt_format:
try:
return time.strftime(timestr, current_format)
except ValueError:
pass
except ValueError:
raise ArgumentError(f'"{timestr}" is an invalid time string') from None
def history(results, key='job', limit=None):
"""Plots run time history of all results. Depending on the argument `job`,
The results are either ordered by runtime (i.e. commit/build time) or
job time (i.e. when the job was run).
Args:
results: List of `result.Result` objects.
key: Either 'job' or 'runtime'.
limit: Optionally limits the number of plotted results to the given
number, i.e. only displays the most recent results. If `None`,
all given results are plotted.
"""
# get date/time either from the runtime (commit/build) or job (when job
# was run)
def get_datetime(result):
if key == 'runtime':
datetime = result.runtime.datetime
elif key == 'job':
datetime = result.datetime
else:
raise ArgumentError('"key" argument must be "runtime" or "job"')
return time.local_time(datetime)
# sort results by desired reference time
results = sorted(results, key=get_datetime)
if limit is not None:
if not isinstance(limit, int) or limit <= 0:
raise ArgumentError('"limit" must be a positive integer')
results = results[-limit:]
data = result.percentiles_by_stencil(results, [0, 25, 50, 75, 100])
dates = [matplotlib.dates.date2num(get_datetime(r)) for r in results]
if len(dates) > len(set(dates)):
logger.warning('Non-unique datetimes in history plot')
locator = matplotlib.dates.AutoDateLocator()
formatter = matplotlib.dates.AutoDateFormatter(locator)
formatter.scaled[1 / 24] = '%y-%m-%d %H:%M'
formatter.scaled[1 / (24 * 60)] = '%y-%m-%d %H:%M'
formatter.scaled[1 / (24 * 60 * 60)] = '%y-%m-%d %H:%M:%S'
if len(data[0]) < 8:
q2, stencil, mint, q1, q3, maxt = zip(*sorted(zip(data[3], data[0], data[1], data[2], data[4], data[5])))
colors = discrete_colors(len(stencil[0]))
fig, ax = plt.subplots(1, 1, figsize=figsize(4, 2))
for color, stencil, mint, q1, q2, q3, maxt in zip(colors, stencil, mint, q1, q2, q3, maxt):
ax.fill_between(dates, mint, maxt, alpha=0.2, color=color)
ax.fill_between(dates, q1, q3, alpha=0.5, color=color)
ax.plot(dates, q2, '|-', label=stencil.title(), color=color)
ax.legend(loc='upper left')
ax.xaxis.set_major_locator(locator)
ax.xaxis.set_major_formatter(formatter)
fig.autofmt_xdate()
fig.tight_layout()
elif len(data[0]) < 20:
# Sort by mean time
three, zero, one, two, four, five = zip(*sorted(zip(data[3], data[0], data[1], data[2], data[4], data[5])))
stencils = numpy.array_split(numpy.array(zero), 4)
mints = numpy.array_split(numpy.array(one), 4)
q1s = numpy.array_split(numpy.array(two), 4)
q2s = numpy.array_split(numpy.array(three), 4)
q3s = numpy.array_split(numpy.array(four), 4)
maxts = numpy.array_split(numpy.array(five), 4)
colors = discrete_colors(len(stencils[0]))
fig, ax = plt.subplots(2, 2, figsize=figsize(4, 2))
axes = itertools.chain(*ax)
for cstencil, cmint, cq1, cq2, cq3, cmaxt, ax in zip(stencils, mints, q1s, q2s, q3s, maxts, axes):
for color, stencil, mint, q1, q2, q3, maxt in zip(colors, cstencil, cmint, cq1, cq2, cq3, cmaxt):
ax.fill_between(dates, mint, maxt, alpha=0.2, color=color)
ax.fill_between(dates, q1, q3, alpha=0.5, color=color)
ax.plot(dates, q2, '|-', label=stencil.title(), color=color)
ax.legend(loc='upper left')
ax.xaxis.set_major_locator(locator)
ax.xaxis.set_major_formatter(formatter)
fig.autofmt_xdate()
fig.tight_layout()
else:
# Sort by mean time
three, zero, one, two, four, five = zip(*sorted(zip(data[3], data[0], data[1], data[2], data[4], data[5])))
stencils = numpy.array_split(numpy.array(zero), 6)
mints = numpy.array_split(numpy.array(one), 6)
q1s = numpy.array_split(numpy.array(two), 6)
q2s = numpy.array_split(numpy.array(three), 6)
q3s = numpy.array_split(numpy.array(four), 6)
maxts = numpy.array_split(numpy.array(five), 6)
colors = discrete_colors(len(stencils[0]))
fig, ax = plt.subplots(3, 2, figsize=figsize(4, 2))
axes = itertools.chain(*ax)
for cstencil, cmint, cq1, cq2, cq3, cmaxt, ax in zip(stencils, mints, q1s, q2s, q3s, maxts, axes):
for color, stencil, mint, q1, q2, q3, maxt in zip(colors, cstencil, cmint, cq1, cq2, cq3, cmaxt):
ax.fill_between(dates, mint, maxt, alpha=0.2, color=color)
ax.fill_between(dates, q1, q3, alpha=0.5, color=color)
ax.plot(dates, q2, '|-', label=stencil.title(), color=color)
ax.legend(loc='upper left')
ax.xaxis.set_major_locator(locator)
ax.xaxis.set_major_formatter(formatter)
fig.autofmt_xdate()
fig.tight_layout()
return fig
def history(results, key='job', limit=None):
"""Plots run time history of all results. Depending on the argument `job`,
The results are either ordered by runtime (i.e. commit/build time) or
job time (i.e. when the job was run).
Args:
results: List of `result.Result` objects.
key: Either 'job' or 'runtime'.
limit: Optionally limits the number of plotted results to the given
number, i.e. only displays the most recent results. If `None`,
all given results are plotted.
"""
# get date/time either from the runtime (commit/build) or job (when job
# was run)
def get_datetime(result):
if key == 'runtime':
datetime = result.runtime.datetime
elif key == 'job':
datetime = result.datetime
else:
raise ArgumentError('"key" argument must be "runtime" or "job"')
return time.local_time(datetime)
# sort results by desired reference time
results = sorted(results, key=get_datetime)
#
if limit is not None:
if not isinstance(limit, int) or limit <= 0:
raise ArgumentError('"limit" must be a positive integer')
results = results[-limit:]
data = result.percentiles_by_stencil(results, [0, 25, 50, 75, 100])
dates = [matplotlib.dates.date2num(get_datetime(r)) for r in results]
if len(dates) > len(set(dates)):
logger.warning('Non-unique datetimes in history plot')
fig, ax = plt.subplots(figsize=figsize(2, 1))
locator = matplotlib.dates.AutoDateLocator()
formatter = matplotlib.dates.AutoDateFormatter(locator)
formatter.scaled[1 / 24] = '%y-%m-%d %H:%M'
formatter.scaled[1 / (24 * 60)] = '%y-%m-%d %H:%M'
formatter.scaled[1 / (24 * 60 * 60)] = '%y-%m-%d %H:%M:%S'
ax.xaxis.set_major_locator(locator)
ax.xaxis.set_major_formatter(formatter)
colors = discrete_colors(len(data[0]))
for color, stencil, mint, q1, q2, q3, maxt in zip(colors, *data):
ax.fill_between(dates, mint, maxt, alpha=0.2, color=color)
ax.fill_between(dates, q1, q3, alpha=0.5, color=color)
ax.plot(dates, q2, '|-', label=stencil.title(), color=color)
ax.legend(loc='upper left')
fig.autofmt_xdate()
fig.tight_layout()
return fig