def ppm_group_algo_contour_plot_helper(test_name, fnames, algos,
granularity=config.PPM_CONTOUR_GRANULARITY,
alpha_start=config.PPM_ALPHA_START, alpha_end=config.PPM_ALPHA_END,
beta_start=config.PPM_BETA_START, beta_end=config.PPM_BETA_END):
if type(fnames) == str:
fnames = [fnames]
alpha_range = (float(alpha_start), float(alpha_end))
beta_range = (float(beta_start), float(beta_end))
granularity = int(granularity)
algos = algos.split(",")
kwargs = contour_settings(config.PPM_GROUP_CONTOUR_DEFAULT_ARGS,
config.PPM_GROUP_CONTOUR_OVERRIDES, test_name, fnames)
fig = plot_contour_base(xlim=beta_range, ylim=alpha_range)
colors = kwargs['colormap'](np.linspace(0, 1, len(algos)))
del kwargs['colormap']
for algo, color in zip(algos, colors):
prior, depth = algo.split("/")
depth = int(depth)
optimum, evals = benchmark.tasks.contour_data(prior, paranoia, depth,
alpha_range, beta_range, granularity, fnames)
label = short_name(config.SHORT_PRIOR_NAME, prior)
plot_optimum(optimum, label=label, color=color)
plot_contour_lines(optimum, evals, colors=color, **kwargs)
return plot.save_figure(fig, general.sanitize_fname(test_name), fnames)
def generate_resource_figure(test, settings, data):
file = settings['file']
algos = settings['algos']
resources, format = process_resource_data(settings, data)
resources = resources[file]
# flatten data
flattened = [confidence_interval(resources[algo], config.RESOURCE_ALPHA) for algo in algos]
x = np.arange(len(algos))
y = [z[0] for z in flattened]
yerr = [z[1] for z in flattened]
xticks = list(map(config.ALGO_ABBREVIATIONS.get, algos))
colors = ppl.brewer2mpl.get_map('Set2', 'qualitative', len(algos)).mpl_colors
if settings['style']:
plot.set_style(settings['style'])
plot.new_figure()
fig, ax = plt.subplots()
rects = ppl.bar(x, y, xticklabels=xticks, yerr=yerr, log=True, grid='y', color=colors)
# Annotate
for rect in rects:
bar_x = rect.get_x() + rect.get_width()/2.
bar_y = rect.get_height()
label = format(bar_y)
plt.annotate(label, xy=(bar_x, bar_y), xytext=(0, 10), textcoords='offset points',
horizontalalignment='center', verticalalignment='bottom')
plt.xlabel('Compressor')
if settings['col'] == 'runtime':
plt.ylabel(r'Runtime (\si{\second})')
elif settings['col'] == 'memory':
plt.ylabel(r'Memory (\si{\byte})')
ax.set_yscale('log', basey=2) # units are in powers of two, so scale should be as well
# hack to make labels fit
ymin, ymax = plt.ylim()
plt.ylim((ymin, ymax*2))
plot.save_figure(fig, 'resources', [test])
def ppm_efficiency_by_depth_helper2(fnames, priors, depths, test_name, opts):
test_files = list(itertools.chain(*config.PPM_EFFICIENCY_BY_DEPTH_FILESETS.values()))
original_sizes = {f : benchmark.tasks.corpus_size(f) for f in fnames}
work = []
for opt, (prior, depth) in zip(opts, itertools.product(priors, depths)):
x, fun, status = opt
a, b = x
work += [benchmark.tasks.my_compressor.s(test_file, paranoia, prior,
['ppm:d={0}:a={1}:b={2}'.format(depth, a, b)])
for test_file in test_files]
raw_res = celery.group(work)().get()
res = {}
for effectiveness, (prior, depth, test_file) in zip(raw_res,
itertools.product(priors, depths, test_files)):
by_prior = res.get(prior, {})
by_depth = by_prior.get(depth, {})
by_depth[test_file] = effectiveness
by_prior[depth] = by_depth
res[prior] = by_prior
fig = plot.new_figure()
colors = ppl.brewer2mpl.get_map('Set2', 'qualitative', len(config.PPM_EFFICIENCY_BY_DEPTH_FILESETS)).mpl_colors
for (name, fileset), color in zip(config.PPM_EFFICIENCY_BY_DEPTH_FILESETS.items(), colors):
for prior in priors:
y = []
for d in depths:
by_file = res[prior][d]
mean = np.mean([by_file[f] / original_sizes[f] * 8 for f in fileset])
y.append(mean)
linestyle = config.PPM_EFFICIENCY_BY_DEPTH_PRIOR_LINESTYLES[prior]
marker = config.PPM_EFFICIENCY_BY_DEPTH_PRIOR_MARKERS[prior]
min_i = np.argmin(y)
markevery = list(range(0, min_i)) + list(range(min_i + 1, len(depths)))
ppl.plot(depths, y, label='{1} on {0}'.format(name, short_name(config.SHORT_PRIOR_NAME, prior)),
color=color, linestyle=linestyle, marker=marker, markevery=markevery)
min_depth = depths[min_i]
min_y = y[min_i]
ppl.plot([min_depth], [min_y], color=color, linestyle='None', marker='D')
plt.xlabel(r'Maximal context depth $d$')
plt.ylabel(r'Compression effectiveness (bits/byte)')
# stretch x-axis slightly so markers are visible
plt.xlim(min(depths) - 0.1, max(depths) + 0.1)
ppl.legend(handlelength=4, # increase length of line segments so that linestyles can be seen
numpoints=1 # but only show marker once
)
return plot.save_figure(fig, test_name, ["dummy"])
def callback(res):
fig = plot_contour_base(xlim=beta_range, ylim=alpha_range)
colors = kwargs['colormap'](np.linspace(0, 1, len(fnames)))
del kwargs['colormap']
for file_res, fname, color in zip(res, fnames, colors):
optimum, evals = file_res
label = short_name(config.SHORT_FILE_NAME, fname)
plot_optimum(optimum, label=label, color=color)
plot_contour_lines(optimum, evals, colors=color, **kwargs)
return plot.save_figure(fig, test_name, fnames)
def ppm_contour_plot_helper(test_name, fnames, prior, depth,
granularity=config.PPM_CONTOUR_GRANULARITY,
alpha_start=config.PPM_ALPHA_START, alpha_end=config.PPM_ALPHA_END,
beta_start=config.PPM_BETA_START, beta_end=config.PPM_BETA_END):
if type(fnames) == str:
fnames = [fnames]
alpha_range = (float(alpha_start), float(alpha_end))
beta_range = (float(beta_start), float(beta_end))
depth = int(depth)
granularity = int(granularity)
optimum, evals = benchmark.tasks.contour_data(prior, paranoia, depth,
alpha_range, beta_range, granularity, fnames)
settings = contour_settings(config.PPM_CONTOUR_DEFAULT_ARGS,
config.PPM_CONTOUR_OVERRIDES, test_name, fnames)
fig = plot_contour_base(xlim=beta_range, ylim=alpha_range)
plot_optimum(optimum[0], **settings['optimum'])
for kwargs in settings['markers'].values():
plot_optimum(**kwargs)
plot_contour_lines(optimum, evals, **settings['lines'])
return plot.save_figure(fig, test_name, fnames)
