Skip to content
Snippets Groups Projects
Commit a94c8df8 authored by Klaus Rabbertz's avatar Klaus Rabbertz
Browse files

Extend runtime plot to use core no. information

parent 0cac6847
Branches
Tags
No related merge requests found
Hide whitespace changes
Inline Side-by-side
tools/plotting/fastnnlo_runtime.py
+ 47
− 25
View file @ a94c8df8
......@@ -120,7 +120,7 @@ def main():
exit(1)
# get all the information from logfiles as dict
# dict contains: runtime, runtime_unit, channel, events
# dict contains: runtimes, runtime_unit, numcores; channel, events
loginformation = get_loginformation(logfiles)
runinformation = get_runinformation(runfiles)
info = {**loginformation, **runinformation}
......@@ -175,10 +175,12 @@ def get_loginformation(files):
# always in hours for simplicity
runtimes = []
numcores = []
unit = 'hours'
for file in files:
runtimes_temp = []
numcores_temp = []
with open(file) as origin:
for line in origin:
......@@ -190,13 +192,20 @@ def get_loginformation(files):
seconds = float(line[3])
runtimes_temp.append(hours + minutes/60 + seconds/3600)
if 'Allocated number of threads' in line:
line = line.split(':')
numcores_temp.append(int(line[1]))
runtimes.append(runtimes_temp[-1])
numcores.append(numcores_temp[-1])
runtimes = np.array(runtimes)
numcores = np.array(numcores)
information = {
'runtime': runtimes,
'runtime_unit': unit
'runtimes': runtimes,
'runtime_unit': unit,
'numcores': numcores
}
return information
......@@ -251,8 +260,9 @@ def get_runinformation(files):
def plot_elapsed_time(infodict, out_path, out_name, formats):
times = infodict['runtime']
unit = infodict['runtime_unit']
times = infodict['runtimes']
unit = infodict['runtime_unit']
cores = infodict['numcores']
channels = infodict['channels']
events = infodict['events']
......@@ -262,6 +272,8 @@ def plot_elapsed_time(infodict, out_path, out_name, formats):
print('fastnnlo_runtime: ERROR! Aborted, no channel info found.')
exit(11)
cputimes = np.multiply(times,cores)
bins = np.linspace(min(times), max(times), 100)
# get relevant values
mean = np.mean(times)
......@@ -271,7 +283,7 @@ def plot_elapsed_time(infodict, out_path, out_name, formats):
# In future?:
# iqd = np.subtract(*np.percentile(times, [75, 25], method='linear'))/2.
CPUtime = np.sum(times)
CPUtime = np.sum(cputimes)
# set figure
fig = plt.figure(figsize=(16, 12))
......@@ -280,26 +292,27 @@ def plot_elapsed_time(infodict, out_path, out_name, formats):
# plot histogram
if len(unique_channels) > 1 or [unique_channels] == 'ALL':
n, batches, _ = ax.hist(times, bins=20, color='deepskyblue', edgecolor='black', label='Total CPU time: {0:0.0f} hours'.format(CPUtime))
n, batches, _ = ax.hist(times, bins=20, color='deepskyblue', edgecolor='black', label='Total CPU time: {0:0.0f} h'.format(CPUtime))
ax.legend(loc='best', fontsize=20)
else:
plt.text
# plot each unique number in different color
for ev_num in set(events):
n, batches, _ = ax.hist(times[events == ev_num], histtype='barstacked', log=True, stacked=True, bins=bins, edgecolor='black', label='# Events: {}'.format(ev_num))
ncore = np.mean(cores[events == ev_num])
n, batches, _ = ax.hist(times[events == ev_num], histtype='barstacked', log=True, stacked=True, bins=bins, edgecolor='black', label='#Events@Cores: {} @ {}'.format(ev_num,ncore))
# plot mean and median
ax.vlines(mean, 0, max(n), colors='red', linestyles='dashed', label=r'Mean: {0:0.1f}$\pm${1:0.1f}'.format(mean, std))
ax.vlines(median, 0, max(n), colors='green', linestyles='dashdot', label=r'Median: {0:0.2f}$\pm${1:0.2f}'.format(median, iqd))
ax.vlines(mean, 0, max(n), colors='red', linestyles='dashed', label=r'Mean run time: {0:0.1f}$\pm${1:0.1f} h'.format(mean, std))
ax.vlines(median, 0, max(n), colors='green', linestyles='dashdot', label=r'Median run time: {0:0.2f}$\pm${1:0.2f} h'.format(median, iqd))
ax.ticklabel_format(axis='x', style='plain', useOffset=False)
ax.legend(title='Total CPU time: {0:0.0f}hours'.format(CPUtime), loc='best', fontsize=20, title_fontsize=20)
ax.legend(title='Total CPU time: {0:0.0f} h'.format(CPUtime), loc='best', fontsize=20, title_fontsize=20)
# finish and save figure
chnlabel = channels[0]
if out_name:
chnlabel = out_name
ax.set_title('Elapsed time of ' + chnlabel + ' production', fontsize=20)
ax.set_xlabel('CPU time [' + unit + ']', horizontalalignment='right', x=1.0, verticalalignment='top', y=1.0, fontsize=20)
ax.set_title('Run times of ' + chnlabel + ' production', fontsize=20)
ax.set_xlabel('Job run time [h]', horizontalalignment='right', x=1.0, verticalalignment='top', y=1.0, fontsize=20)
ax.set_ylabel('# jobs', horizontalalignment='right', x=1.0, verticalalignment='top', y=1.0, fontsize=20, labelpad=20)
ax.set_yscale('log')
ax.tick_params(axis='both', which='major', labelsize=20)
......@@ -323,36 +336,44 @@ def plot_events_per_hour(infodict, out_path, out_name, formats):
# get input
channels = infodict['channels']
events = infodict['events']
times = infodict['runtime']
cores = infodict['numcores']
times = infodict['runtimes']
unit = infodict['runtime_unit']
# prepare input
cputimes = np.multiply(times,cores)
unique_channels = set(channels)
if len(unique_channels) == 0:
print('fastnnlo_runtime: ERROR! Aborted, no channel info found.')
exit(11)
eph = []
ncr = []
# for i, time in enumerate(cputimes): # This would be per hour & core!
for i, time in enumerate(times):
eph.append(float(events[i])/time)
ncr.append(float(cores[i]))
ephchn = []
ncrchn = []
for i, val in enumerate(eph):
for j, chn in enumerate(_channels):
if channels[i] == chn:
ephchn.append([val, j])
ncrchn.append([ncr[i], j])
ephchn = np.array(ephchn)
ncrchn = np.array(ncrchn)
masks = []
for i, chn in enumerate(_channels):
masks.append(ephchn[:,1] == i)
# get relevant values
mean = np.mean(eph)
std = np.std(eph)
median = np.median(eph)
iqd = np.subtract(*np.percentile(eph, [75, 25], interpolation='linear'))/2.
ephmin = np.min(eph)
ephmax = np.max(eph)
logbins = np.geomspace(ephmin, ephmax, 100)
CPUtime = np.sum(times)
mean = np.mean(eph)
std = np.std(eph)
median = np.median(eph)
iqd = np.subtract(*np.percentile(eph, [75, 25], interpolation='linear'))/2.
ephmin = np.min(eph)
ephmax = np.max(eph)
logbins = np.geomspace(ephmin, ephmax, 100)
CPUtime = np.sum(times)
# create figure
fig = plt.figure(figsize=(16, 12))
......@@ -360,11 +381,12 @@ def plot_events_per_hour(infodict, out_path, out_name, formats):
# plot (multistack-)histogram
evrs = []
ncrs = []
for chn in _channels:
if chn in unique_channels:
evrs.append(ephchn[masks[_channel_number[chn]]][:,0])
ncrs.append(ncrchn[masks[_channel_number[chn]]][:,0])
if len(unique_channels) == 1:
# plot each unique number in different color
for ev_num in set(events):
n, batches, _ = ax.hist(evrs[0][events == ev_num], histtype='barstacked', log=True, stacked=True, bins=50, edgecolor='black', label='# Events: {}'.format(ev_num))
......@@ -376,9 +398,9 @@ def plot_events_per_hour(infodict, out_path, out_name, formats):
else:
chnlab = []
chncol = []
for chn in _channels:
for i, chn in enumerate(_channels):
if chn in unique_channels:
chnlab.append(chn)
chnlab.append(chn+' @ '+str(np.mean(ncrs[i]))+' core(s)')
chncol.append(_channel_colors[_channel_number[chn]])
n, batches, _ = ax.hist(evrs, histtype='barstacked', log=True, stacked=True, bins=logbins, edgecolor='black', color=chncol, label=chnlab)
ax.set_xlim(0.9*ephmin, 1.1*ephmax)
......
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Please register or to comment