def test_get_color_pool(self):
"""Test grabbing list of good colors to use."""
obs = get_color_pool()
self.assertEqual(len(obs), 27)
obs2 = get_color_pool()
self.assertFloatEqual(obs, obs2)
self.assertFalse(obs is obs2)
def _collate_cluster_pcoa_plot_data(coords_f, map_f, category):
pc_data = parse_coords(coords_f)
coords_d = dict(zip(pc_data[0], pc_data[1]))
map_data = parse_mapping_file(map_f)
full_map_data = [map_data[1]]
full_map_data.extend(map_data[0])
sid_map = group_by_field(full_map_data, category)
sorted_states = sorted(sid_map.keys())
color_pool = get_color_pool()
if len(sorted_states) > len(color_pool):
raise ValueError("Not enough colors to uniquely color sample "
"groups.")
results = []
for state, color in zip(sorted_states,
color_pool[:len(sorted_states)]):
sids = sid_map[state]
xs = [coords_d[sid][0] for sid in sids]
ys = [coords_d[sid][1] for sid in sids]
results.append((xs, ys, color, state))
return results
def create_simulated_data_plots(analysis_type, in_dir, workflow):
"""Create plots of sample size vs effect size/p-val for each dissim.
Plots will be placed directly under in_dir and will be named according to
the following convention:
<study>_<category>_<depth>_<metric>.pdf
"""
for study in workflow:
study_dir = join(in_dir, study)
num_trials = workflow[study]['num_sim_data_trials']
methods = workflow[study]['methods']
if Best() in methods:
methods.remove(Best())
if MantelCorrelogram() in methods:
methods.remove(MantelCorrelogram())
num_methods = len(methods)
num_rows = max(num_methods, len(workflow[study]['pcoa_dissim']) + 1)
# test stat, p-val, legend/PCoA.
num_cols = 3
for depth in workflow[study]['depths']:
depth_dir = join(study_dir, '%d' % depth[0])
data_type_dir = join(depth_dir, 'simulated')
for category in workflow[study]['categories']:
category_dir = join(data_type_dir, category[0])
# metric -> Figure
figs = {}
for metric in workflow[study]['metrics']:
figs[metric[0]] = figure(num=None, figsize=(20, 20),
facecolor='w', edgecolor='k')
for method_idx, method in enumerate(methods):
# metric ->
# dissim -> {
# 'sample_sizes': list,
# 'effect_sizes': list of lists, one for each size,
# 'p_vals' -> list of lists, one for each size
# }
plots_data = defaultdict(lambda:
defaultdict(lambda: defaultdict(list)))
for trial_num in range(num_trials):
trial_num_dir = join(category_dir, '%d' % trial_num)
for samp_size in workflow[study]['sample_sizes']:
samp_size_dir = join(trial_num_dir,
'%d' % samp_size)
for d in workflow[study]['plot_dissim']:
dissim_dir = join(samp_size_dir, repr(d))
for metric in workflow[study]['metrics']:
metric_dir = join(dissim_dir, metric[0])
method_dir = join(metric_dir,
method.DirectoryName)
results_fp = join(method_dir,
'%s_results.txt' %
method.ResultsName)
effect_size, p_val = method.parse(
open(results_fp, 'U'))
if samp_size not in plots_data[metric[0]][d]['sample_sizes']:
plots_data[metric[0]][d]['sample_sizes'].append(samp_size)
plots_data[metric[0]][d]['effect_sizes'].append([])
plots_data[metric[0]][d]['p_vals'].append([])
samp_size_idx = plots_data[metric[0]][d]['sample_sizes'].index(samp_size)
plots_data[metric[0]][d]['effect_sizes'][samp_size_idx].append(effect_size)
plots_data[metric[0]][d]['p_vals'][samp_size_idx].append(p_val)
for metric in workflow[study]['metrics']:
fig = figs[metric[0]]
metric_plots_data = plots_data[metric[0]]
# plot_num is 1-based indexing.
plot_num = method_idx * num_cols + 1
ax1 = fig.add_subplot(num_rows, num_cols, plot_num)
ax2 = fig.add_subplot(num_rows, num_cols, plot_num + 1)
color_pool = get_color_pool()
min_dissim = min(metric_plots_data.keys())
max_dissim = max(metric_plots_data.keys())
legend_labels = []
legend_lines = []
for d, plot_data in sorted(metric_plots_data.items()):
avg_effect_sizes, std_effect_sizes, avg_p_vals, std_p_vals = \
_compute_plot_data_statistics(plot_data, num_trials)
color = color_pool.pop(0)
label = 'd=%r' % d
if d == 0.0:
label += ' (actual data)'
#elif d == max_dissim:
# label += ' (neg. control)'
legend_labels.append(label)
legend_lines.append(Line2D([0, 1], [0, 0],
color=color, linewidth=2))
# Make the actual data plot a bit thicker than
# the rest.
if d == 0.0:
line_width = 3
else:
line_width = 1
# Plot test statistics.
ax1.errorbar(plot_data['sample_sizes'],
avg_effect_sizes, yerr=std_effect_sizes,
color=color, label=label,
linewidth=line_width, fmt='-')
# Plot p-values.
_, _, barlinecols = ax2.errorbar(
plot_data['sample_sizes'], avg_p_vals,
yerr=std_p_vals, color=color, label=label,
linewidth=line_width, linestyle='--')
barlinecols[0].set_linestyles('dashed')
ax1.set_xscale('log', nonposx='clip', basex=2)
ax1.xaxis.set_major_formatter(FormatStrFormatter('%d'))
ax2.set_xscale('log', nonposx='clip', basex=2)
ax2.xaxis.set_major_formatter(FormatStrFormatter('%d'))
ax2.set_yscale('log', nonposy='clip')
x_label = 'Number of samples'
ax1.set_xlabel(x_label)
ax2.set_xlabel(x_label)
ax1.set_ylabel('%s (%s)' % (method.DisplayName,
method.StatDisplayName))
ax2.set_ylabel('p-value')
min_x = min(workflow[study]['sample_sizes'])
max_x = max(workflow[study]['sample_sizes'])
ax1.set_xlim(min_x - 0.5, max_x)
ax2.set_xlim(min_x - 0.5, max_x)
for ax_idx, ax in enumerate((ax1, ax2)):
panel_idx = method_idx * 2 + ax_idx
panel_label = get_panel_label(panel_idx)
xmin = ax.get_xlim()[0]
ymin, ymax = ax.get_ylim()
yrange = ymax - ymin
# Not sure why the math isn't working out for the
# p-value plots...
if ax is ax1:
factor = 0.05
else:
factor = 0.60
ax.text(xmin, ymax + (factor * yrange),
'(%s)' % panel_label)
if method_idx == 0:
ax3 = fig.add_subplot(num_rows, num_cols,
plot_num + 2, frame_on=False)
ax3.get_xaxis().set_visible(False)
ax3.get_yaxis().set_visible(False)
start_panel_label = get_panel_label(0)
end_panel_label = \
get_panel_label(num_methods * 2 - 1)
if analysis_type == 'gradient':
loc='center'
elif analysis_type == 'cluster':
loc='center left'
assert len(legend_lines) == len(workflow[study]['plot_dissim'])
assert len(legend_labels) == len(workflow[study]['plot_dissim'])
legend_title = (' Legend (Panels %s-%s)\nd = '
'"noise" introduced to samples' % (
start_panel_label, end_panel_label))
ax3.legend(legend_lines, legend_labels, ncol=1,
title=legend_title, loc=loc, fancybox=True,
shadow=True)
for metric in workflow[study]['metrics']:
fig = figs[metric[0]]
# Plot PCoA in last column of figure.
plot_pcoa(analysis_type, fig, category_dir, workflow[study],
category, metric, num_rows, num_cols, num_methods)
fig.tight_layout(pad=5.0, w_pad=2.0, h_pad=2.0)
fig.savefig(join(in_dir, '%s_%s_%d_%s.pdf' % (study,
category[0], depth[0], metric[0])), format='pdf')
fig.savefig(join(in_dir, '%s_%s_%d_%s.png' % (study,
category[0], depth[0], metric[0])), format='png',
dpi=100)
