def compare(self, others, scores, dtype=np.float16, plot=False):
result0 = self.compute(scores, dtype=dtype)
if not isiterable(others):
others = [others]
result_grid = []
for other in others:
result1 = other.compute(scores, dtype=dtype)
if plot:
from matplotlib import pyplot as plt
from palettable import colorbrewer
colors = colorbrewer.get_map("Set1", "qualitative", 9).mpl_colors
result_row = {}
for score_name, scores0 in result0.iteritems():
scores1 = result1[score_name]
auc_score = dist_auc(scores0, scores1)
result_row[score_name] = auc_score
if plot:
scores0p = [x for x in scores0 if not np.isnan(x)]
scores1p = [x for x in scores1 if not np.isnan(x)]
hmin0, hmax0 = minmaxr(scores0p)
hmin1, hmax1 = minmaxr(scores1p)
bins = np.linspace(min(hmin0, hmin1), max(hmax0, hmax1), 50)
plt.hist(scores0p, bins, alpha=0.5, label="0", color=colors[0], edgecolor="none")
plt.hist(scores1p, bins, alpha=0.5, label="1", color=colors[1], edgecolor="none")
plt.legend(loc="upper right")
plt.title("%s: AUC=%.4f" % (score_name, auc_score))
plt.show()
result_grid.append(result_row)
return result_grid
def auc_xscaled(xs, ys):
"""AUC score scaled to fill x interval
"""
xmin, xmax = minmaxr(xs)
denom = float(xmax - xmin)
xs_corr = [(x - xmin) / denom for x in xs]
return auc(xs_corr, ys)
def test_minmaxr_2(self):
arr = []
amin, amax = minmaxr(arr)
self.assertTrue(np.isinf(amin))
self.assertTrue(np.isinf(amin))
self.assertGreater(amin, amax)
def test_minmaxr_1(self):
arr = [2, 3, 4]
amin, amax = minmaxr(arr)
self.assertAlmostEqual(2.0, amin)
self.assertAlmostEqual(4.0, amax)
def create_plots(args, df):
import jinja2
import matplotlib.pyplot as plt
from palettable import colorbrewer
from matplotlib.font_manager import FontProperties
fontP = FontProperties()
fontP.set_size("xx-small")
# groups = df.set_index(args.x_axis).groupby([args.group_by])
groups = df.groupby([args.group_by])
metrics = list(set(args.metrics) & set(df.keys()))
colors = take(
len(metrics),
cycle(chain(colorbrewer.qualitative.Dark2_8.mpl_colors, colorbrewer.qualitative.Set2_8.mpl_colors)),
)
template_loader = jinja2.FileSystemLoader(os.path.join(args.output, ".."))
template_env = jinja2.Environment(loader=template_loader)
template_interactive = template_env.get_template("template_fig_interactive.html")
template_static = template_env.get_template("template_fig_static.html")
table_interactive = []
table_static = []
for group_name, group in groups:
# always sort by X values
group = group.sort([args.x_axis])
if args.fig_title is None:
fig_title = "%s=%s" % (args.group_by, group_name)
else:
fig_title = args.fig_title
# compute AUC scores
ys = []
for metric, color in zip(metrics, colors):
series = group[metric]
score = auc_xscaled(group[args.x_axis].values, series.values)
label = "%s (%.4f)" % (metric, score)
ys.append((score, metric, label, color))
ys.sort(reverse=True)
lbls_old, lbls_new, colors = zip(*ys)[1:4]
group = (
group[[args.x_axis] + list(lbls_old)].set_index(args.x_axis).rename(columns=dict(zip(lbls_old, lbls_new)))
)
# create plots
fig, ax = plt.subplots()
group.plot(ax=ax, title=fig_title, color=list(colors))
ax.set_xlim(*minmaxr(group.index.values))
ax.set_ylim(0.4, 1.0)
ax.legend(loc=args.legend_loc, prop=fontP)
fig_name = "fig-%s.%s" % (group_name, args.fig_format)
fig_path = os.path.join(args.output, fig_name)
csv_name = "fig-%s.csv" % group_name
csv_path = os.path.join(args.output, csv_name)
group.to_csv(csv_path)
table_interactive.append((csv_name, args.x_axis, "%s=%s" % (args.group_by, group_name)))
table_static.append(fig_name)
fig.savefig(fig_path, format=args.fig_format)
plt.close(fig)
with open(os.path.join(args.output, "fig_interactive.html"), "w") as fh:
fh.write(template_interactive.render(table=table_interactive))
with open(os.path.join(args.output, "fig_static.html"), "w") as fh:
fh.write(template_static.render(table=table_static))
