def clustdm(cmeth, dm, nclust, cargs={}):
if cmeth == 'tree':
t = clust.dtree(dm, **cargs)
clsts = t.cut(nclust)
elif cmeth == 'med':
clsts = clust.mediods(dm, nclust, **cargs)
else:
raise StandardError("don't know clustering method %s" % cmeth)
return clsts
def clustm(d, nc, mode='med', npass=1000):
for c in dconds(d):
dm = d[c]['dm']
if mode == 'med':
cd = clust.kmedoids(dm, nc, npass)[0]
elif mode == 'tree':
cd = clust.dtree(dm).cut(nc)
cids = list(set(cd))
cid = []
for id in cd:
cid.append(cids.index(id))
d[c]['cid'] = cid
def show_cell_clust(cells, cn, mode='vdps', q=6000, fig=1,
save="", suf=''):
"""
cell: i, mode: DistMode, q: DistQ, fig: i, save: DirName, suf: s
-> None (draws in matplotlib figure fig, may write a file)
Shows the clustering tree (as show_tree) for each stimulus in each condition,
and also the clustering diagram. Draws a figure. If save is non-empty, then
also writes a file in that directory, named: "c%i_clust_%s%i%s.png" % (cell,
mode, int(q),suf)
"""
d = cells[cn]
f = plt.figure(fig)
plt.clf()
plt.figtext(0, .98, "%s, mode=%s, q=%.2g" % (cn, mode, q), backgroundcolor='white')
labs = stimnames(d)
a1 = dist_discrim_incond(d['cond1'], mode, q, len(labs))
a2 = dist_discrim_incond(d['cond2'], mode, q, len(labs))
sb = plt.subplot(121)
plt.title("Condition 1")
im = vis.dot2img(clust.tree2dot(clust.dtree(a1), labs))
plt.imshow(im)
sb.yaxis.set_visible(False)
sb.xaxis.set_visible(False)
sb = plt.subplot(122)
plt.title("Condition 2")
im = vis.dot2img(clust.tree2dot(clust.dtree(a2), labs))
plt.imshow(im)
sb.yaxis.set_visible(False)
sb.xaxis.set_visible(False)
f.canvas.draw()
if save:
save = os.path.expanduser(save)
fn = os.path.join(save, "%s_clust_%s%i%s.png" % (cn, mode, int(q),
suf))
plt.savefig(fn)
def _tsort(a, l):
"""
a: DistMat(N), l: N-[ of s -> DistMat(N), N-[ of s
creates a new matrix and list of names sorted using tree clustering
"""
t = clust.dtree(a)
co = clust.treesort(t)
l2 = [l[i] for i in co]
a2 = np.zeros_like(a)
for i in range(a.shape[0]):
for j in range(a.shape[1]):
a2[i, j] = a[co[i], co[j]]
return (a2, l2)
def noise_tree(cond, nbs=20, noise=.1, mode='vdps', q=6000, N=14,
ctree=True):
"""
Cond: IntIO
Construct a set of <nbs> different clustering trees using a distance matrix
calculated with distance function <mode> with parameter <q>. The trees
differ because in each of the nbs trials, the DM is perturbed with
independent Gaussian noise drawn from N(0, S), where S = <noise>*DM.std().
consensus tree of trees generated by adding noise values to the distance
matrix. These values are drawn from a N(0, sd) where sd is noise*dm.std. <N>
is the total number of stimuli available (used by the internal call to
dist_discrim_incond.
If <ctree> is True (default), returns a majority rule consensus tree
across the various clusterings. Otherwise, returns a list of the
cluster trees
If 'mode' is "random_*", then the dm used is a random matrix where *
specifies the distribution (it is passed as the first argument to
random_dm). Can be used for testing.
"""
if mode.startswith('random_'):
dm = random_dm(mode[7:], N)
else:
dm = dist_discrim_incond(cond, mode, q, N)
nsd = noise * dm.std()
trees = []
for _ in range(nbs):
dm2 = dm + np.random.normal(0, nsd, dm.shape)
dm2 = np.maximum(dm2, 0)
trees.append(clust.dtree(dm2))
if ctree:
return clust.mrtree(trees)
else:
return trees
def run(self, pars, out, messages):
dm = np.array(pars['dm'])
t = clust.dtree(dm)
out['tree'] = clust.tree2tup(t)