def plot_timing(charges, tolerance):
ixcha = all_charges.index(charges)
ixtol = all_tolerances.index(tolerance)
# look for the minimal timing
minimal_time = None
for method in all_methods:
methodtimes = timing[method][ixcha,ixtol,:]
if all(isnan(methodtimes)): continue
i = 0
while (isnan(methodtimes[i])): i += 1
serial_time = all_cores[i] * methodtimes[i]
if minimal_time is None or serial_time < minimal_time:
minimal_time = serial_time
loglog(all_cores, minimal_time/array(all_cores), 'k:')
loglog(all_cores, 10.*minimal_time/array(all_cores), 'k:')
loglog(all_cores, 100.*minimal_time/array(all_cores), 'k:')
for method in all_methods:
data = timing[method]
if not all(isnan(data[ixcha,ixtol,:])):
loglog(all_cores, data[ixcha,ixtol,:],
label=method, **benchmarks.fmt(method))
xlim((1, charges/200.))
xscale('log', basex=2)
xlabel('\#Cores $P$')
ylabel('Time $t$ [s]')
def plot_timing(charges, tolerance):
ixcha = all_charges.index(charges)
ixtol = all_tolerances.index(tolerance)
# look for the minimal timing
minimal_time = None
for method in all_methods:
methodtimes = timing[method][ixcha, ixtol, :]
if all(isnan(methodtimes)): continue
i = 0
while (isnan(methodtimes[i])):
i += 1
serial_time = all_cores[i] * methodtimes[i]
if minimal_time is None or serial_time < minimal_time:
minimal_time = serial_time
loglog(all_cores, minimal_time / array(all_cores), 'k:')
loglog(all_cores, 10. * minimal_time / array(all_cores), 'k:')
loglog(all_cores, 100. * minimal_time / array(all_cores), 'k:')
for method in all_methods:
data = timing[method]
if not all(isnan(data[ixcha, ixtol, :])):
loglog(all_cores,
data[ixcha, ixtol, :],
label=method,
**benchmarks.fmt(method))
xlim((1, charges / 200.))
xscale('log', basex=2)
xlabel('\#Cores $P$')
ylabel('Time $t$ [s]')
def plot_efficiency(charges, tolerance):
ixcha = all_charges.index(charges)
ixtol = all_tolerances.index(tolerance)
# look for the minimal timing
minimal_time = None
for method in all_methods:
methodtimes = timing[method][ixcha,ixtol,:]
if all(isnan(methodtimes)): continue
i = 0
while (isnan(methodtimes[i])): i += 1
serial_time = all_cores[i] * methodtimes[i]
if minimal_time is None or serial_time < minimal_time:
minimal_time = serial_time
if minimal_time is None:
print "No data for charges={} tolerance={}"\
.format(charges, tolerance)
return
for method in all_methods:
data = timing[method]
if not all(isnan(data[ixcha,ixtol,:])):
semilogx(all_cores, minimal_time/(data[ixcha,ixtol,:]*all_cores),
label=method, **benchmarks.fmt(method))
# legend()
xlim((1, charges/200.))
xscale('log', basex=2)
xlabel(r'\#Cores $P$')
ylabel(r'Relative Parallel Efficiency $e(P)$')
def plot_efficiency(charges, tolerance):
ixcha = all_charges.index(charges)
ixtol = all_tolerances.index(tolerance)
# look for the minimal timing
minimal_time = None
for method in all_methods:
methodtimes = timing[method][ixcha, ixtol, :]
if all(isnan(methodtimes)): continue
i = 0
while (isnan(methodtimes[i])):
i += 1
serial_time = all_cores[i] * methodtimes[i]
if minimal_time is None or serial_time < minimal_time:
minimal_time = serial_time
if minimal_time is None:
print "No data for charges={} tolerance={}"\
.format(charges, tolerance)
return
for method in all_methods:
data = timing[method]
if not all(isnan(data[ixcha, ixtol, :])):
semilogx(all_cores,
minimal_time / (data[ixcha, ixtol, :] * all_cores),
label=method,
**benchmarks.fmt(method))
# legend()
xlim((1, charges / 200.))
xscale('log', basex=2)
xlabel(r'\#Cores $P$')
ylabel(r'Relative Parallel Efficiency $e(P)$')
def plot_efficiency(testcase, charges, tolerance):
ixcha = all_charges.index(charges)
ixtol = all_tolerances.index(tolerance)
methods = timing.keys()
methods.sort()
minimal_time = None
# look for the minimal timing
for method in methods:
methodtimes = timing[method][ixcha, ixtol, :]
if all(isnan(methodtimes)): continue
i = 0
while (isnan(methodtimes[i])):
i += 1
serial_time = all_cores[i] * methodtimes[i]
if minimal_time is None or serial_time < minimal_time:
minimal_time = serial_time
if minimal_time is None:
print "No data for charges={} tolerance={}".format(charges, tolerance)
return
for method in methods:
data = timing[method]
if not all(isnan(data[ixcha, ixtol, :])):
semilogx(all_cores,
minimal_time / (data[ixcha, ixtol, :] * all_cores),
label=method,
**benchmarks.fmt(method))
xlim((1, charges / 200.))
xlabel('#cores')
ylabel('Efficiency')
def plot_efficiency(charges, tolerance):
ixcha = all_charges.index(charges)
ixtol = all_tolerances.index(tolerance)
methods = timing.keys()
methods.sort()
minimal_time = None
# look for the minimal timing
for method in methods:
methodtimes = timing[method][ixcha,ixtol,:]
if all(isnan(methodtimes)): continue
i = 0
while (isnan(methodtimes[i])): i += 1
serial_time = all_cores[i] * methodtimes[i]
if minimal_time is None or serial_time < minimal_time:
minimal_time = serial_time
if minimal_time is None:
print "No data for charges={} tolerance={}".format(charges, tolerance)
return
for method in methods:
data = timing[method]
if not all(isnan(data[ixcha,ixtol,:])):
semilogx(all_cores, minimal_time/(data[ixcha,ixtol,:]*all_cores),
label=method, **benchmarks.fmt(method))
# legend()
xlim((1, charges/200.))
xlabel('#cores')
ylabel('Efficiency')
def plot_timing(charges, tolerance):
ixcha = all_charges.index(charges)
ixtol = all_tolerances.index(tolerance)
methods = timing.keys()
methods.sort()
for method in methods:
data = timing[method]
if not all(isnan(data[ixcha,ixtol,:])):
loglog(all_cores, data[ixcha,ixtol,:],
label=method, **benchmarks.fmt(method))
xlim((1, charges/200.))
xlabel('#cores')
ylabel('Time [s]')
def plot_timing(charges, tolerance):
ixcha = all_charges.index(charges)
ixtol = all_tolerances.index(tolerance)
methods = timing.keys()
methods.sort()
for method in methods:
data = timing[method]
if not all(isnan(data[ixcha,ixtol,:])):
plt.loglog(all_cores, data[ixcha,ixtol,:],
label=method, **benchmarks.fmt(method))
plt.legend()
plt.xlim((1, charges/200.))
plt.xlabel('#cores')
plt.ylabel('Time [s]')
all_cores = data['cores']
all_tolerances = data['tolerances']
all_charges = data['charges']
timing = data['timing']
cores = 1
ixcor = all_cores.index(cores)
tolerance = 1e-3
ixtol = all_tolerances.index(tolerance)
figure('Complexity')
for method in data['methods']:
time_per_particle = \
timing[method][:, ixtol, ixcor] / all_charges
loglog(all_charges, time_per_particle,
label=method, **benchmarks.fmt(method))
xlabel(r'\#Charges')
ylabel(r'Time $t$/\#Charges [s]')
create_legend()
savefig('complexity.pdf')
####################
# ACCURACY
####################
data = benchmarks.read('accuracy.xml')
all_methods = data['methods']
all_cores = data['cores']
all_tolerances = data['tolerances']
all_charges = data['charges']
timing = data['timing']
all_charges = data['charges']
timing = data['timing']
cores = 1
ixcor = all_cores.index(cores)
tolerance = 1e-3
ixtol = all_tolerances.index(tolerance)
figure('Complexity')
for method in data['methods']:
time_per_particle = \
timing[method][:, ixtol, ixcor] / all_charges
loglog(all_charges,
time_per_particle,
label=method,
**benchmarks.fmt(method))
xlabel(r'\#Charges')
ylabel(r'Time $t$/\#Charges [s]')
create_legend()
savefig('complexity.pdf')
####################
# ACCURACY
####################
data = benchmarks.read('accuracy.xml')
all_methods = data['methods']
all_cores = data['cores']
all_tolerances = data['tolerances']
all_charges = data['charges']
timing = data['timing']
