def test_cuts(self):
labels = cut_straight(self.dendrogram)
self.assertEqual(len(set(labels)), 2)
labels = cut_straight(self.dendrogram, n_clusters=5)
self.assertEqual(len(set(labels)), 5)
labels = cut_balanced(self.dendrogram)
self.assertEqual(len(set(labels)), 21)
labels, new_dendrogram = cut_balanced(self.dendrogram,
return_dendrogram=True)
self.assertEqual(len(set(labels)), 21)
self.assertTupleEqual(new_dendrogram.shape, (20, 4))
def test_options(self):
labels = cut_straight(self.dendrogram, threshold=0.5)
self.assertEqual(len(set(labels)), 7)
labels = cut_straight(self.dendrogram, n_clusters=3, threshold=0.5)
self.assertEqual(len(set(labels)), 3)
labels = cut_straight(self.dendrogram, sort_clusters=False)
self.assertEqual(len(set(labels)), 2)
labels = cut_balanced(self.dendrogram,
max_cluster_size=2,
sort_clusters=False)
self.assertEqual(len(set(labels)), 21)
labels = cut_balanced(self.dendrogram, max_cluster_size=10)
self.assertEqual(len(set(labels)), 5)
def filter(self, n_clusters):
try:
self.dendrogram
except AttributeError:
var_exists = False
else:
var_exists = True
if not var_exists:
self._cluster()
self.df['cluster_id'] = cut_straight(self.dendrogram, n_clusters)
return Draw.MolsToGridImage(
self.df.drop_duplicates(subset='cluster_id', keep='first')['mols'])
def cluster(self, method, n_clust=None, threshold=None):
"""Cuts the dendrogram and returns cluster IDs. Straight cuts can either
set a defined number of clusters, or alternatively set a distance threshold.
Cluster sizes can vary widely.
Balanced cuts respect a maximum cluster size. The number of clusters is determined
on the fly. """
if self.verbose:
print(f'clustering with a {method} cut')
if method == 'straight':
if n_clust is not None and threshold is not None:
raise ValueError('Straight cut takes only one of n_clusters or threshold, not both.')
self.clusters = cut_straight(self.dendrogram, n_clust, threshold)
elif method == 'balanced':
if n_clust is None:
raise ValueError('Must set maximum cluster size (n_clust) for balanced_cut')
self.clusters = cut_balanced(self.dendrogram, n_clust)
else:
print('Choose \"straight\" or \"balanced\"')
def svg_dendrogram_top(dendrogram, names, width, height, margin, margin_text,
scale, line_width, n_clusters, color, font_size,
reorder, rotate_names):
"""Dendrogram as SVG image with root on top."""
# scaling
height *= scale
width *= scale
# positioning
labels = cut_straight(dendrogram, n_clusters, return_dendrogram=False)
index = get_index(dendrogram, reorder)
n = len(index)
unit_height = height / dendrogram[-1, 2]
unit_width = width / n
height_basis = margin + height
position = {
index[i]: (margin + i * unit_width, height_basis)
for i in range(n)
}
label = {i: l for i, l in enumerate(labels)}
width += 2 * margin
height += 2 * margin
if names is not None:
text_length = np.max(np.array([len(str(name)) for name in names]))
height += text_length * font_size * .5 + margin_text
svg = """<svg width="{}" height="{}" xmlns="http://www.w3.org/2000/svg">""".format(
width, height)
# text
if names is not None:
for i in range(n):
x, y = position[i]
x -= margin_text
y += margin_text
if rotate_names:
svg += """<text x="{}" y="{}" transform="rotate(60, {}, {})" font-size="{}">{}</text>""" \
.format(x, y, x, y, font_size, str(names[i]))
else:
y += margin_text
svg += """<text x="{}" y="{}" font-size="{}">{}</text>""" \
.format(x, y, font_size, str(names[i]))
# tree
for t in range(n - 1):
i = int(dendrogram[t, 0])
j = int(dendrogram[t, 1])
x1, y1 = position.pop(i)
x2, y2 = position.pop(j)
l1 = label.pop(i)
l2 = label.pop(j)
if l1 == l2:
line_color = STANDARD_COLORS[l1 % len(STANDARD_COLORS)]
else:
line_color = color
x = .5 * (x1 + x2)
y = height_basis - dendrogram[t, 2] * unit_height
svg += """<path stroke-width="{}" stroke="{}" d="M {} {} {} {}" />"""\
.format(line_width, line_color, x1, y1, x1, y)
svg += """<path stroke-width="{}" stroke="{}" d="M {} {} {} {}" />"""\
.format(line_width, line_color, x2, y2, x2, y)
svg += """<path stroke-width="{}" stroke="{}" d="M {} {} {} {}" />"""\
.format(line_width, line_color, x1, y, x2, y)
position[n + t] = (x, y)
label[n + t] = l1
svg += '</svg>'
return svg
def adjacency_clusters(adj,n_clusters=None,threshold=None):
print('Clustering molecules...')
ward = Ward()
dendrogram = ward.fit_transform(adj)
labels = cut_straight(dendrogram,n_clusters=n_clusters,threshold=threshold)
return labels
def svg_dendrogram_left(dendrogram, names, width, height, margin, margin_text, scale, line_width, n_clusters,
color, colors, font_size, reorder):
"""Dendrogram as SVG image with root on left side."""
# scaling
height *= scale
width *= scale
# positioning
labels = cut_straight(dendrogram, n_clusters, return_dendrogram=False)
index = get_index(dendrogram, reorder)
n = len(index)
unit_height = height / n
unit_width = width / dendrogram[-1, 2]
width_basis = width + margin
position = {index[i]: (width_basis, margin + i * unit_height) for i in range(n)}
label = {i: l for i, l in enumerate(labels)}
width += 2 * margin
height += 2 * margin
if names is not None:
text_length = np.max(np.array([len(str(name)) for name in names]))
width += text_length * font_size * .5 + margin_text
svg = """<svg width="{}" height="{}" xmlns="http://www.w3.org/2000/svg">""".format(width, height)
# text
if names is not None:
for i in range(n):
x, y = position[i]
x += margin_text
y += unit_height / 3
text = str(names[i]).replace('&', ' ')
svg += """<text x="{}" y="{}" font-size="{}">{}</text>""" \
.format(x, y, font_size, text)
# tree
for t in range(n - 1):
i = int(dendrogram[t, 0])
j = int(dendrogram[t, 1])
x1, y1 = position.pop(i)
x2, y2 = position.pop(j)
l1 = label.pop(i)
l2 = label.pop(j)
if l1 == l2:
line_color = colors[l1 % len(colors)]
else:
line_color = color
y = .5 * (y1 + y2)
x = width_basis - dendrogram[t, 2] * unit_width
svg += """<path stroke-width="{}" stroke="{}" d="M {} {} {} {}" />"""\
.format(line_width, line_color, x1, y1, x, y1)
svg += """<path stroke-width="{}" stroke="{}" d="M {} {} {} {}" />"""\
.format(line_width, line_color, x2, y2, x, y2)
svg += """<path stroke-width="{}" stroke="{}" d="M {} {} {} {}" />"""\
.format(line_width, line_color, x, y1, x, y2)
position[n + t] = (x, y)
label[n + t] = l1
svg += '</svg>'
return svg
