def generate_histogram(subgroups_to_sses_to_n_count, tname, file_name):
columns_to_data = {'subgroup': [], tname: [], 'count': []}
max_count = 0
for subgroup, sses_to_n_count in subgroups_to_sses_to_n_count.items():
for ss, n_count in sses_to_n_count.items():
columns_to_data['subgroup'].append(subgroup)
columns_to_data[tname].append(ss)
columns_to_data['count'].append(n_count)
if n_count > max_count:
max_count = n_count
r_columns_to_data = {
'subgroup':
ro.FactorVector(columns_to_data['subgroup'],
levels=ro.StrVector(
_sort_subgroup(set(columns_to_data['subgroup'])))),
tname:
ro.StrVector(columns_to_data[tname]),
'count':
ro.IntVector(columns_to_data['count'])
}
df = ro.DataFrame(r_columns_to_data)
max_count = int(max_count / 1000 * 1000 + 1000)
histogram_file_path = os.path.join(OUTPUT_PATH, file_name)
logging.debug(
str.format("The Data Frame for file {}: \n{}", histogram_file_path,
df))
grdevices.png(file=histogram_file_path, width=1200, height=800)
gp = ggplot2.ggplot(df)
pp = gp + \
ggplot2.aes_string(x='subgroup', y='count', fill=tname) + \
ggplot2.geom_bar(position="dodge",width=0.8, stat="identity") + \
ggplot2.theme_bw() + \
ggplot2.theme_classic() + \
ggplot2.theme(**{'legend.title': ggplot2.element_blank()}) + \
ggplot2.theme(**{'legend.text': ggplot2.element_text(size=40)}) + \
ggplot2.theme(**{'axis.text.x': ggplot2.element_text(size=40,angle=45)}) + \
ggplot2.theme(**{'axis.text.y': ggplot2.element_text(size=40)}) + \
ggplot2.scale_y_continuous(expand=ro.IntVector([0, 0]),
limits=ro.IntVector([0, max_count])) + \
ggplot2.geom_text(ggplot2.aes_string(label='count'), size=6, angle=35, hjust=-0.1,
position=ggplot2.position_dodge(width=0.8),
vjust=-0.2)
pp.plot()
logging.info(str.format("Output step3 file {}", histogram_file_path))
grdevices.dev_off()
def bargraph_language(cfg, values):
r = robjects.r
for lang in cfg.languages:
times = []
varss = []
probs = []
ses = []
for prob in cfg.problems:
for var in cfg.variations:
# we use the pretty names to make the
varss.append(pretty_varis[var])
probs.append(prob)
data = FloatVector(values[prob][var][lang][0])
times.append(r['mean'](data)[0])
t_result = r['t.test'](data, **{
" conf.level": 0.999
}).rx('conf.int')[0]
ses.append((t_result[1] - t_result[0]) / 2)
r.pdf('bargraph-executiontime-lang-' + lang + '.pdf',
height=pdf_height(),
width=pdf_width())
df = robjects.DataFrame({
'Variation': StrVector(varss),
'Problem': StrVector(probs),
'Time': FloatVector(times),
'SE': FloatVector(ses)
})
limits = ggplot2.aes(ymax='Time + SE', ymin='Time - SE')
dodge = ggplot2.position_dodge(width=0.9)
gp = ggplot2.ggplot(df)
pp = gp + \
ggplot2.aes_string (x='Problem', y='Time', fill='Variation') + \
ggplot2.geom_bar (position='dodge', stat='identity') + \
ggplot2.geom_errorbar (limits, position=dodge, width=0.25) + \
ggplot2_options () + \
ggplot2_colors () + \
robjects.r('scale_x_discrete(limits=c("randmat", "thresh", "winnow", "outer", "product", "chain"))') +\
robjects.r('ylab("Execution time (in seconds)")')
pp.plot()
r['dev.off']()
def bargraph_language (cfg, values):
r = robjects.r
for lang in cfg.languages:
times = []
varss = []
probs = []
ses = []
for prob in cfg.problems:
for var in cfg.variations:
# we use the pretty names to make the
varss.append (pretty_varis [var])
probs.append (prob)
data = FloatVector (values[prob][var][lang][0])
times.append (r['mean'] (data)[0])
t_result = r['t.test'] (data,
**{" conf.level": 0.999}).rx ('conf.int')[0]
ses.append ((t_result[1] - t_result[0])/2)
r.pdf ('bargraph-executiontime-lang-' + lang + '.pdf',
height=pdf_height (), width=pdf_width ())
df = robjects.DataFrame({'Variation': StrVector (varss),
'Problem': StrVector (probs),
'Time' : FloatVector (times),
'SE' : FloatVector (ses)
})
limits = ggplot2.aes (ymax = 'Time + SE', ymin = 'Time - SE')
dodge = ggplot2.position_dodge (width=0.9)
gp = ggplot2.ggplot (df)
pp = gp + \
ggplot2.aes_string (x='Problem', y='Time', fill='Variation') + \
ggplot2.geom_bar (position='dodge', stat='identity') + \
ggplot2.geom_errorbar (limits, position=dodge, width=0.25) + \
ggplot2_options () + \
ggplot2_colors () + \
robjects.r('scale_x_discrete(limits=c("randmat", "thresh", "winnow", "outer", "product", "chain"))') +\
robjects.r('ylab("Execution time (in seconds)")')
pp.plot ()
r['dev.off']()
def generate_step3_9_n_count_histogram(place_type_pos_type_to_count,
file_name):
columns_to_data = {'place': [], 'pos': [], 'count': []}
max_count = 0
for place_pos_type, n_count in place_type_pos_type_to_count.items():
place_type, pos_type = place_pos_type.split('_')
columns_to_data['place'].append(place_type)
columns_to_data['pos'].append(pos_type)
columns_to_data['count'].append(n_count)
if n_count > max_count:
max_count = n_count
r_columns_to_data = {
'place': ro.StrVector(columns_to_data['place']),
'pos': ro.StrVector(columns_to_data['pos']),
'count': ro.IntVector(columns_to_data['count'])
}
df = ro.DataFrame(r_columns_to_data)
if max_count > 1000:
max_count = int(max_count / 1000 * 1000 + 1000)
else:
max_count = int(max_count / 100 * 100 + 100)
histogram_file_path = os.path.join(OUTPUT_PATH, file_name)
logging.debug(
str.format("The Data Frame for file {}: \n{}", histogram_file_path,
df))
grdevices.png(file=histogram_file_path, width=1024, height=512)
gp = ggplot2.ggplot(df)
pp = gp + \
ggplot2.aes_string(x='pos', y='count', fill='place') + \
ggplot2.geom_bar(position="dodge", stat="identity") + \
ggplot2.theme_bw() + \
ggplot2.theme_classic() + \
ggplot2.theme(**{'axis.text.x': ggplot2.element_text(size=35)}) + \
ggplot2.theme(**{'axis.text.y': ggplot2.element_text(size=35)}) + \
ggplot2.scale_y_continuous(expand=ro.IntVector([0, 0]),
limits=ro.IntVector([0, max_count])) + \
ggplot2.geom_text(ggplot2.aes_string(label='count'),
position=ggplot2.position_dodge(width=0.8), size=10, angle=35, hjust=-0.2,
vjust=-0.5)
pp.plot()
logging.info(str.format("Output step3 file {}", histogram_file_path))
grdevices.dev_off()
def rest():
df = q1_median_q3_rep_wide
pops = ["pdc", "dc-cd11b", "dc-cd8a"]
stats_l = []
for stat, (popa, popb) in product(["Q1", "median", "Q3"],
product(pops, pops)):
print(stat, popa, popb)
popa = "hsc"
popb = "pdc"
stat = "median"
mw_u, pvalue = scipy.stats.mannwhitneyu(
[0.8, 0.81, 0.79],
[0.4, 0.39, 0.41],
# df.query("Population == @popa")[stat].to_numpy(),
# df.query("Population == @popb")[stat].to_numpy(),
use_continuity=True,
alternative="two-sided",
)
pvalue
stats_l.append([stat, popa, popb, mw_u, pvalue])
stats_df = pd.DataFrame(stats_l).set_axis(
["stat", "popA", "popB", "U", "pvalue"], axis=1)
kruskal_format_means = pd.pivot(
q1_median_q3_rep_wide.query("Population in @pops"),
index="Population",
columns="Replicate",
values="mean",
)
import scikit_posthocs
stat, p_value = scipy.stats.kruskal(
*[kruskal_format_means.loc[pop].to_numpy() for pop in pops], )
dunn_res_df = scikit_posthocs.posthoc_dunn(
kruskal_format_means.to_numpy(),
p_adjust='fdr_bh',
sort=True,
)
stat, pvalue = scipy.stats.f_oneway(
*[kruskal_format_means.loc[pop].to_numpy() for pop in pops], )
import statsmodels
df = kruskal_format_means.stack().reset_index()
kruskal_format_means
res = statsmodels.stats.multicomp.pairwise_tukeyhsd(
df[0], df['Population'].to_numpy(), alpha=0.05)
res.pvalues
res.summary()
# wilcox.test(c(0.8, 0.79, 0.81), c(0.4, 0.39, 0.41), paired=F, exact=F)
plot_pops = ["pdc", "dc-cd8a", "dc-cd11b"]
results_dir = "/icgc/dkfzlsdf/analysis/hs_ontogeny/notebook-data/gNs4xcMJscaLLwlt"
point_plot_quartiles_png = results_dir + "/point-plot-quartiles.png"
q1_median_q3_rep_wide
ggplot_data = (
q1_median_q3_rep_long.query("Population in @plot_pops").sort_values(
"value",
ascending=False,
).groupby(["Population", "stat"]).apply(
lambda df: df.assign(group_order=np.arange(1, df.shape[0] + 1))))
g = (gg.ggplot(ggplot_data) + gg.aes_string(
x="Population", y="value", group="group_order", color="stat") +
gg.geom_point(position=gg.position_dodge(width=0.5), size=1) +
mh_rpy2_styling.gg_paper_theme + gg.labs(y='Methylation (%)', x=''))
a = 3
rpy2_utils.image_png2(g, (ut.cm(6), ut.cm(6)))
ut.save_and_display(
g,
png_path=point_plot_quartiles_png,
# additional_formats=tuple(),
height=ut.cm(6),
width=ut.cm(6),
)
q1_median_q3_rep_wide
g = (
gg.ggplot(
q1_median_q3_rep_wide.query("Population in @plot_pops").assign(
sample=lambda df: df["Population"].astype(str) + df[
"Replicate"].astype(str))) + gg.geom_boxplot(
gg.aes_string(
x="Population",
fill="Population",
group="sample",
lower="Q1",
upper="Q3",
middle="median",
ymin="min1",
ymax="max99",
# position=gg.position_dodge(width=0.5),
),
stat="identity",
)
# + mh_rpy2_styling.gg_paper_theme
+ gg.theme(axis_text_x=gg.element_text(angle=90, hjust=1)) +
gg.scale_fill_brewer(guide=False))
a = 3
ut.save_and_display(
g,
png_path=point_plot_quartiles_png,
additional_formats=tuple(),
height=ut.cm(6),
width=ut.cm(7),
)
# image_png2(g, (ut.cm(12), ut.cm(12)))
beta_values.loc[:, ("hsc", "1")]
def bargraph_variation():
r = robjects.r
for var in variations:
# each variation gets plot
values = []
# normalized values
nvalues = []
langs = []
probs = []
for prob in problems:
# aggregate by problems
lvalues = []
for lang in languages:
# each problem displays a list of language times for that problem
langs.append(pretty_langs[lang])
probs.append(prob)
value = 0
try:
value = result[lang][prob][var]
except KeyError:
print "Warning: no value for:"
print(lang, prob, var)
value = 0 # FIXME to account for missing seq-version of Erlang
# for the expert times, add expert and non-expert times together
if var.startswith('expert'):
try:
value = value + result[lang][prob][var.replace(
'expert', '')]
except KeyError:
pass
lvalues.append(value)
values.extend(lvalues)
lmin = min([x for x in lvalues if x != 0])
nvalues.extend([(lambda x: x / lmin)(la) for la in lvalues])
# plot histogram of actual times
r.pdf('bargraph-codingtime-var-' + var + '.pdf',
height=pdf_height(),
width=pdf_width())
df = robjects.DataFrame({
'Language': StrVector(langs),
'Problem': StrVector(probs),
'Time': FloatVector(values),
})
dodge = ggplot2.position_dodge(width=0.9)
gp = ggplot2.ggplot(df)
pp = gp + \
ggplot2.aes_string (x='Problem', y='Time', fill='Language') + \
ggplot2.geom_bar (position='dodge', stat='identity') + \
ggplot2_options () + \
ggplot2_colors () + \
robjects.r('scale_x_discrete(limits=c("randmat", "thresh", "winnow", "outer", "product", "chain"))') +\
robjects.r('ylab("Coding time (in minutes)")')
pp.plot()
# plot histogram of times normalized with respect to fastest time for a problem
r.pdf('bargraph-codingtime-var-norm-' + var + '.pdf',
height=pdf_height(),
width=pdf_width())
df = robjects.DataFrame({
'Language': StrVector(langs),
'Problem': StrVector(probs),
'Time': FloatVector(nvalues),
})
dodge = ggplot2.position_dodge(width=0.9)
gp = ggplot2.ggplot(df)
pp = gp + \
ggplot2.aes_string (x='Problem', y='Time', fill='Language') + \
ggplot2.geom_bar (position='dodge', stat='identity') + \
ggplot2_options () + \
ggplot2_colors () + \
robjects.r('scale_x_discrete(limits=c("randmat", "thresh", "winnow", "outer", "product", "chain"))') +\
robjects.r('ylab("Coding time (normalized to fastest)")')
pp.plot()
r['dev.off']()
def line_plot (cfg, var, control, change_name, changing, selector, base_selector, basis):
speedups = []
thrds = []
changes = []
lowers = []
uppers = []
for n in cfg.threads:
probs.append ('ideal')
langs.append ('ideal')
speedups.append (n)
thrds.append (n)
changes.append ('ideal')
lowers.append (n)
uppers.append (n)
for c in changing:
sel = selector (c)
# sequential base
base = FloatVector (base_selector(c))
# base with p = 1
base_p1 = FloatVector (sel(1))
# use fastest sequential program
if basis == 'fastest' and mean (base_p1) < mean(base):
base = base_p1
elif basis == 'seq':
pass
elif basis == 'p1':
base = base_p1
for n in cfg.threads:
ntimes = FloatVector (sel(n))
# ratio confidence interval
labels = ['Base'] * r.length(base)[0] + ['N']*r.length (ntimes)[0]
df = DataFrame ({'Times': base + ntimes,
'Type': StrVector(labels)})
ratio_test = r['pairwiseCI'] (r('Times ~ Type'), data=df,
control='N',
method='Param.ratio',
**{'var.equal': False,
'conf.level': 0.999})[0][0]
lowers.append (ratio_test[1][0])
uppers.append (ratio_test[2][0])
mn = mean (ntimes)
speedups.append (mean(base) / mn)
# plot slowdowns
#speedups.append (-mn/base)#(base / mn)
thrds.append (n)
if change_name == 'Language':
changes.append (pretty_langs [c])
else:
changes.append (c)
df = DataFrame ({'Speedup': FloatVector (speedups),
'Threads': IntVector (thrds),
change_name: StrVector (changes),
'Lower': FloatVector (lowers),
'Upper': FloatVector (uppers)
})
ideal_changing = ['ideal']
if change_name == 'Language':
ideal_changing.extend ([pretty_langs [c] for c in changing])
else:
ideal_changing.extend (changing)
legendVec = IntVector (range (len (ideal_changing)))
legendVec.names = StrVector (ideal_changing)
gg = ggplot2.ggplot (df)
limits = ggplot2.aes (ymax = 'Upper', ymin = 'Lower')
dodge = ggplot2.position_dodge (width=0.9)
pp = gg + \
ggplot2.geom_line() + ggplot2.geom_point(size=3) +\
ggplot2.aes_string(x='Threads', y='Speedup',
group=change_name, color=change_name,
shape=change_name) + \
ggplot2.scale_shape_manual(values=legendVec) + \
ggplot2.geom_errorbar (limits, width=0.25) + \
ggplot2_options () + \
ggplot2_colors () + \
ggplot2.opts (**{'axis.title.x' : ggplot2.theme_text(family = 'serif', face = 'bold', size = 15, vjust=-0.2)}) + \
robjects.r('ylab("Speedup")') + \
robjects.r('xlab("Cores")')
# ggplot2.xlim (min(threads), max(threads)) + ggplot2.ylim(min(threads), max(threads)) +\
pp.plot()
r['dev.off']()
def as_dataframe (cfg, results, basis):
r = robjects.r
varis = []
langs = []
probs = []
times = []
threads = []
# speedups, with upper and lower bounds below
speedups = []
speedup_lowers = []
speedup_uppers = []
ses = [] # standard errors
mems = [] # memory usage
langs_ideal = list (cfg.languages)
langs_ideal.append ('ideal')
probs_ideal = list (cfg.problems)
probs_ideal.append ('ideal')
for var in cfg.variations:
for lang in langs_ideal: # cfg.languages:
for prob in probs_ideal: # cfg.problems:
for thread in cfg.threads:
if lang == 'ideal' and prob == 'ideal':
continue
elif lang == 'ideal' or prob == 'ideal':
varis.append (var)
langs.append (pretty_langs[lang])
probs.append (prob)
threads.append (thread)
speedups.append (thread)
speedup_lowers.append (thread)
speedup_uppers.append (thread)
times.append (0)
ses.append(0)
mems.append (0)
continue
varis.append (var) # pretty_varis [var])
langs.append (pretty_langs [lang])
probs.append (prob)
threads.append (thread)
if var.find('seq') >= 0:
thread = cfg.threads[-1]
vals = FloatVector (results[thread][prob][var][lang][0])
time = mean (vals)
times.append (time)
#
# time confidence interval
#
t_result = r['t.test'] (FloatVector(vals),
**{" conf.level": 0.999}).rx ('conf.int')[0]
ses.append ((t_result[1] - t_result[0])/2)
#
# memory usage
#
mem_filename = get_mem_output (lang, prob, var)
with open (mem_filename, 'r') as mem_file:
mem = mem_file.readline()
mems.append (float (mem))
# we include dummy data for the sequential case to avoid the
# speedup calculation below
if var.find('seq') >= 0:
speedups.append (1)
speedup_lowers.append (1)
speedup_uppers.append (1)
continue
#
# speedup values and confidence intervals
#
seq_vals = results[cfg.threads[-1]][prob][var.replace ('par', 'seq')][lang][0]
# sequential base
base = FloatVector (seq_vals)
# base with p = 1
base_p1 = FloatVector (results[1][prob][var][lang][0])
# use fastest sequential program
if basis == 'fastest' and mean (base_p1) < mean(base):
base = base_p1
elif basis == 'seq':
pass
elif basis == 'p1':
base = base_p1
labels = ['Base'] * r.length(base)[0] + ['N']*r.length (vals)[0]
df = DataFrame ({'Times': base + vals,
'Type': StrVector(labels)})
ratio_test = r['pairwiseCI'] (r('Times ~ Type'), data=df,
control='N',
method='Param.ratio',
**{'var.equal': False})[0][0]
speedups.append (mean(base) / time)
speedup_lowers.append (ratio_test[1][0])
speedup_uppers.append (ratio_test[2][0])
df = robjects.DataFrame({'Language': StrVector (langs),
'Problem': StrVector (probs),
'Variation' : StrVector (varis),
'Threads': IntVector (threads),
'Time': FloatVector (times),
'SE': FloatVector (ses),
'Speedup': FloatVector (speedups),
'SpeedupLower': FloatVector (speedup_lowers),
'SpeedupUpper': FloatVector (speedup_uppers),
'Mem' : FloatVector (mems)
})
r.assign ('df', df)
r ('save (df, file="performance.Rda")')
# reshape the data to make variation not a column itself, but a part of
# the other columns describe ie, time, speedup, etc.
#
# also, remove the 'ideal' problem as we don't want it in this plot.
df = r('''
redf = reshape (df,
timevar="Variation",
idvar = c("Language","Problem","Threads"),
direction="wide")
redf$Problem <- factor(redf$Problem, levels = c("randmat","thresh","winnow","outer","product","chain"))
redf[which(redf$Problem != "ideal"),]
''')
r.pdf ('speedup-expertpar-all.pdf',
height=6.5, width=10)
change_name = 'Language'
legendVec = IntVector (range (len (langs_ideal)))
legendVec.names = StrVector (langs_ideal)
gg = ggplot2.ggplot (df)
limits = ggplot2.aes (ymax = 'SpeedupUpper.expertpar', ymin = 'SpeedupLower.expertpar')
dodge = ggplot2.position_dodge (width=0.9)
pp = gg + \
ggplot2.geom_line() + ggplot2.geom_point(size=2.5) +\
robjects.r('scale_color_manual(values = c("#ffcb7e", "#1da06b", "#b94646", "#00368a", "#CCCCCC"))') +\
ggplot2.aes_string(x='Threads', y='Speedup.expertpar',
group=change_name, color=change_name,
shape=change_name) + \
ggplot2.geom_errorbar (limits, width=0.25) + \
ggplot2.opts (**{'axis.title.x' : ggplot2.theme_text(family = 'serif', face = 'bold', size = 10, vjust=-0.2),
'axis.title.y' : ggplot2.theme_text(family = 'serif', face = 'bold', size = 10, angle=90, vjust=0.2),
'axis.text.x' : ggplot2.theme_text(family = 'serif', size = 10),
'axis.text.y' : ggplot2.theme_text(family = 'serif', size = 10),
'legend.title' : ggplot2.theme_text(family = 'serif', face = 'bold', size = 10),
'legend.text' : ggplot2.theme_text(family = 'serif', size = 10),
'strip.text.x' : ggplot2.theme_text(family = 'serif', size = 10),
'aspect.ratio' : 1,
}) + \
robjects.r('ylab("Speedup")') + \
robjects.r('xlab("Number of cores")') + \
ggplot2.facet_wrap ('Problem', nrow = 2)
pp.plot()
r['dev.off']()
def bargraph_variation (cfg, values):
r = robjects.r
for var in cfg.variations:
# each variation gets plot
avgs = []
ses = []
# normalized values
navgs = []
nses = []
langs = []
probs = []
for prob in cfg.problems:
# aggregate by problems
lavgs = []
lses = []
for lang in cfg.languages:
# each problem displays a list of language times for that problem
data = FloatVector (values[prob][var][lang][0])
langs.append (pretty_langs [lang])
probs.append (prob)
mean = r['mean'] (data)[0]
lavgs.append (mean)
t_result = r['t.test'] (data, **{"conf.level": 0.999}).rx ('conf.int')[0]
lses.append ((t_result[1] - t_result[0])/2)
avgs.extend (lavgs)
ses.extend (lses)
lmin = min (lavgs)
navgs.extend ([la/lmin for la in lavgs])
nses.extend ([ls/lmin for ls in lses])
df = robjects.DataFrame({'Language': StrVector (langs),
'Problem': StrVector (probs),
'Time' : FloatVector (avgs),
'SE' : FloatVector (ses),
'NormTime' : FloatVector (navgs),
'NormSE' : FloatVector (nses),
'TimeLabel' : StrVector ([str(round(time, 1)) + "s" for time in avgs])
})
# plot histogram of actual times
r.pdf ('bargraph-executiontime-var-' + var + '.pdf', height=pdf_height (), width=pdf_width ())
limits = ggplot2.aes (ymax = 'Time + SE', ymin = 'Time - SE')
dodge = ggplot2.position_dodge (width=0.9)
gp = ggplot2.ggplot (df)
pp = gp + \
ggplot2.aes_string (x='Problem', y='Time', fill='Language') + \
ggplot2.geom_bar (position='dodge', stat='identity') + \
ggplot2.geom_errorbar (limits, position=dodge, width=0.25) + \
ggplot2_options () + \
ggplot2_colors () + \
robjects.r('scale_x_discrete(limits=c("randmat", "thresh", "winnow", "outer", "product", "chain"))') +\
robjects.r('ylab("Execution time (in seconds)")')
pp.plot ()
# plot histogram of times normalized with respect to fastest time for a problem
r.pdf ('bargraph-executiontime-var-norm-' + var + '.pdf', height=pdf_height (), width=pdf_width ())
limits = ggplot2.aes (ymax = 'NormTime + NormSE', ymin = 'NormTime - NormSE')
dodge = ggplot2.position_dodge (width=0.9)
gp = ggplot2.ggplot (df)
pp = gp + \
ggplot2.aes_string (x='Problem', y='NormTime', fill='Language') + \
ggplot2.geom_bar (position='dodge', stat='identity') + \
ggplot2.geom_errorbar (limits, position=dodge, width=0.25) +\
ggplot2_options () + \
ggplot2_colors () + \
robjects.r('scale_x_discrete(limits=c("randmat", "thresh", "winnow", "outer", "product", "chain"))') +\
robjects.r('ylab("Execution time (normalized to fastest)")')
#ggplot2.geom_text(data=df,
# mapping = ggplot2.aes_string (x='Problem',
# y='NormTime + NormSE + 0.1',
# label='TimeLabel')
pp.plot ()
r['dev.off']()
def line_plot(cfg, var, control, change_name, changing, selector,
base_selector, basis):
speedups = []
thrds = []
changes = []
lowers = []
uppers = []
for n in cfg.threads:
probs.append('ideal')
langs.append('ideal')
speedups.append(n)
thrds.append(n)
changes.append('ideal')
lowers.append(n)
uppers.append(n)
for c in changing:
sel = selector(c)
# sequential base
base = FloatVector(base_selector(c))
# base with p = 1
base_p1 = FloatVector(sel(1))
# use fastest sequential program
if basis == 'fastest' and mean(base_p1) < mean(base):
base = base_p1
elif basis == 'seq':
pass
elif basis == 'p1':
base = base_p1
for n in cfg.threads:
ntimes = FloatVector(sel(n))
# ratio confidence interval
labels = ['Base'] * r.length(base)[0] + ['N'] * r.length(ntimes)[0]
df = DataFrame({'Times': base + ntimes, 'Type': StrVector(labels)})
ratio_test = r['pairwiseCI'](r('Times ~ Type'),
data=df,
control='N',
method='Param.ratio',
**{
'var.equal': False,
'conf.level': 0.999
})[0][0]
lowers.append(ratio_test[1][0])
uppers.append(ratio_test[2][0])
mn = mean(ntimes)
speedups.append(mean(base) / mn)
# plot slowdowns
#speedups.append (-mn/base)#(base / mn)
thrds.append(n)
if change_name == 'Language':
changes.append(pretty_langs[c])
else:
changes.append(c)
df = DataFrame({
'Speedup': FloatVector(speedups),
'Threads': IntVector(thrds),
change_name: StrVector(changes),
'Lower': FloatVector(lowers),
'Upper': FloatVector(uppers)
})
ideal_changing = ['ideal']
if change_name == 'Language':
ideal_changing.extend([pretty_langs[c] for c in changing])
else:
ideal_changing.extend(changing)
legendVec = IntVector(range(len(ideal_changing)))
legendVec.names = StrVector(ideal_changing)
gg = ggplot2.ggplot(df)
limits = ggplot2.aes(ymax='Upper', ymin='Lower')
dodge = ggplot2.position_dodge(width=0.9)
pp = gg + \
ggplot2.geom_line() + ggplot2.geom_point(size=3) +\
ggplot2.aes_string(x='Threads', y='Speedup',
group=change_name, color=change_name,
shape=change_name) + \
ggplot2.scale_shape_manual(values=legendVec) + \
ggplot2.geom_errorbar (limits, width=0.25) + \
ggplot2_options () + \
ggplot2_colors () + \
ggplot2.opts (**{'axis.title.x' : ggplot2.theme_text(family = 'serif', face = 'bold', size = 15, vjust=-0.2)}) + \
robjects.r('ylab("Speedup")') + \
robjects.r('xlab("Cores")')
# ggplot2.xlim (min(threads), max(threads)) + ggplot2.ylim(min(threads), max(threads)) +\
pp.plot()
r['dev.off']()
def as_dataframe(cfg, results, basis):
r = robjects.r
varis = []
langs = []
probs = []
times = []
threads = []
# speedups, with upper and lower bounds below
speedups = []
speedup_lowers = []
speedup_uppers = []
ses = [] # standard errors
mems = [] # memory usage
langs_ideal = list(cfg.languages)
langs_ideal.append('ideal')
probs_ideal = list(cfg.problems)
probs_ideal.append('ideal')
for var in cfg.variations:
for lang in langs_ideal: # cfg.languages:
for prob in probs_ideal: # cfg.problems:
for thread in cfg.threads:
if lang == 'ideal' and prob == 'ideal':
continue
elif lang == 'ideal' or prob == 'ideal':
varis.append(var)
langs.append(pretty_langs[lang])
probs.append(prob)
threads.append(thread)
speedups.append(thread)
speedup_lowers.append(thread)
speedup_uppers.append(thread)
times.append(0)
ses.append(0)
mems.append(0)
continue
varis.append(var) # pretty_varis [var])
langs.append(pretty_langs[lang])
probs.append(prob)
threads.append(thread)
if var.find('seq') >= 0:
thread = cfg.threads[-1]
vals = FloatVector(results[thread][prob][var][lang][0])
time = mean(vals)
times.append(time)
#
# time confidence interval
#
t_result = r['t.test'](FloatVector(vals), **{
" conf.level": 0.999
}).rx('conf.int')[0]
ses.append((t_result[1] - t_result[0]) / 2)
#
# memory usage
#
mem_filename = get_mem_output(lang, prob, var)
with open(mem_filename, 'r') as mem_file:
mem = mem_file.readline()
mems.append(float(mem))
# we include dummy data for the sequential case to avoid the
# speedup calculation below
if var.find('seq') >= 0:
speedups.append(1)
speedup_lowers.append(1)
speedup_uppers.append(1)
continue
#
# speedup values and confidence intervals
#
seq_vals = results[cfg.threads[-1]][prob][var.replace(
'par', 'seq')][lang][0]
# sequential base
base = FloatVector(seq_vals)
# base with p = 1
base_p1 = FloatVector(results[1][prob][var][lang][0])
# use fastest sequential program
if basis == 'fastest' and mean(base_p1) < mean(base):
base = base_p1
elif basis == 'seq':
pass
elif basis == 'p1':
base = base_p1
labels = ['Base'
] * r.length(base)[0] + ['N'] * r.length(vals)[0]
df = DataFrame({
'Times': base + vals,
'Type': StrVector(labels)
})
ratio_test = r['pairwiseCI'](r('Times ~ Type'),
data=df,
control='N',
method='Param.ratio',
**{
'var.equal': False
})[0][0]
speedups.append(mean(base) / time)
speedup_lowers.append(ratio_test[1][0])
speedup_uppers.append(ratio_test[2][0])
df = robjects.DataFrame({
'Language': StrVector(langs),
'Problem': StrVector(probs),
'Variation': StrVector(varis),
'Threads': IntVector(threads),
'Time': FloatVector(times),
'SE': FloatVector(ses),
'Speedup': FloatVector(speedups),
'SpeedupLower': FloatVector(speedup_lowers),
'SpeedupUpper': FloatVector(speedup_uppers),
'Mem': FloatVector(mems)
})
r.assign('df', df)
r('save (df, file="performance.Rda")')
# reshape the data to make variation not a column itself, but a part of
# the other columns describe ie, time, speedup, etc.
#
# also, remove the 'ideal' problem as we don't want it in this plot.
df = r('''
redf = reshape (df,
timevar="Variation",
idvar = c("Language","Problem","Threads"),
direction="wide")
redf$Problem <- factor(redf$Problem, levels = c("randmat","thresh","winnow","outer","product","chain"))
redf[which(redf$Problem != "ideal"),]
''')
r.pdf('speedup-expertpar-all.pdf', height=6.5, width=10)
change_name = 'Language'
legendVec = IntVector(range(len(langs_ideal)))
legendVec.names = StrVector(langs_ideal)
gg = ggplot2.ggplot(df)
limits = ggplot2.aes(ymax='SpeedupUpper.expertpar',
ymin='SpeedupLower.expertpar')
dodge = ggplot2.position_dodge(width=0.9)
pp = gg + \
ggplot2.geom_line() + ggplot2.geom_point(size=2.5) +\
robjects.r('scale_color_manual(values = c("#ffcb7e", "#1da06b", "#b94646", "#00368a", "#CCCCCC"))') +\
ggplot2.aes_string(x='Threads', y='Speedup.expertpar',
group=change_name, color=change_name,
shape=change_name) + \
ggplot2.geom_errorbar (limits, width=0.25) + \
ggplot2.opts (**{'axis.title.x' : ggplot2.theme_text(family = 'serif', face = 'bold', size = 10, vjust=-0.2),
'axis.title.y' : ggplot2.theme_text(family = 'serif', face = 'bold', size = 10, angle=90, vjust=0.2),
'axis.text.x' : ggplot2.theme_text(family = 'serif', size = 10),
'axis.text.y' : ggplot2.theme_text(family = 'serif', size = 10),
'legend.title' : ggplot2.theme_text(family = 'serif', face = 'bold', size = 10),
'legend.text' : ggplot2.theme_text(family = 'serif', size = 10),
'strip.text.x' : ggplot2.theme_text(family = 'serif', size = 10),
'aspect.ratio' : 1,
}) + \
robjects.r('ylab("Speedup")') + \
robjects.r('xlab("Number of cores")') + \
ggplot2.facet_wrap ('Problem', nrow = 2)
pp.plot()
r['dev.off']()
def bargraph_variation(cfg, values):
r = robjects.r
for var in cfg.variations:
# each variation gets plot
avgs = []
ses = []
# normalized values
navgs = []
nses = []
langs = []
probs = []
for prob in cfg.problems:
# aggregate by problems
lavgs = []
lses = []
for lang in cfg.languages:
# each problem displays a list of language times for that problem
data = FloatVector(values[prob][var][lang][0])
langs.append(pretty_langs[lang])
probs.append(prob)
mean = r['mean'](data)[0]
lavgs.append(mean)
t_result = r['t.test'](data, **{
"conf.level": 0.999
}).rx('conf.int')[0]
lses.append((t_result[1] - t_result[0]) / 2)
avgs.extend(lavgs)
ses.extend(lses)
lmin = min(lavgs)
navgs.extend([la / lmin for la in lavgs])
nses.extend([ls / lmin for ls in lses])
df = robjects.DataFrame({
'Language':
StrVector(langs),
'Problem':
StrVector(probs),
'Time':
FloatVector(avgs),
'SE':
FloatVector(ses),
'NormTime':
FloatVector(navgs),
'NormSE':
FloatVector(nses),
'TimeLabel':
StrVector([str(round(time, 1)) + "s" for time in avgs])
})
# plot histogram of actual times
r.pdf('bargraph-executiontime-var-' + var + '.pdf',
height=pdf_height(),
width=pdf_width())
limits = ggplot2.aes(ymax='Time + SE', ymin='Time - SE')
dodge = ggplot2.position_dodge(width=0.9)
gp = ggplot2.ggplot(df)
pp = gp + \
ggplot2.aes_string (x='Problem', y='Time', fill='Language') + \
ggplot2.geom_bar (position='dodge', stat='identity') + \
ggplot2.geom_errorbar (limits, position=dodge, width=0.25) + \
ggplot2_options () + \
ggplot2_colors () + \
robjects.r('scale_x_discrete(limits=c("randmat", "thresh", "winnow", "outer", "product", "chain"))') +\
robjects.r('ylab("Execution time (in seconds)")')
pp.plot()
# plot histogram of times normalized with respect to fastest time for a problem
r.pdf('bargraph-executiontime-var-norm-' + var + '.pdf',
height=pdf_height(),
width=pdf_width())
limits = ggplot2.aes(ymax='NormTime + NormSE',
ymin='NormTime - NormSE')
dodge = ggplot2.position_dodge(width=0.9)
gp = ggplot2.ggplot(df)
pp = gp + \
ggplot2.aes_string (x='Problem', y='NormTime', fill='Language') + \
ggplot2.geom_bar (position='dodge', stat='identity') + \
ggplot2.geom_errorbar (limits, position=dodge, width=0.25) +\
ggplot2_options () + \
ggplot2_colors () + \
robjects.r('scale_x_discrete(limits=c("randmat", "thresh", "winnow", "outer", "product", "chain"))') +\
robjects.r('ylab("Execution time (normalized to fastest)")')
#ggplot2.geom_text(data=df,
# mapping = ggplot2.aes_string (x='Problem',
# y='NormTime + NormSE + 0.1',
# label='TimeLabel')
pp.plot()
r['dev.off']()
