def plot_margins(serie, weights, clfs, importances=None):
global directory
if directory is None:
directory = prepare_directory()
try:
from sklearn import tree
except ImportError:
return
feature_names = ["f{} ({}, {})".format(i // 2, i, '-' if (i % 2) == 0 else '+') for i in range(2 * len(serie))]
out_str = io.StringIO()
for clf in clfs:
tree.export_graphviz(clf, out_file=out_str, feature_names=feature_names)
print("\n\n", file=out_str)
with open(str(directory / "tree.dot"), "w") as ofile:
print(out_str.getvalue(), file=ofile)
dtww.plot_margins(serie, weights, filename=str(directory / "margins.png"), importances=importances)
def plotsuperinstancemargins(clustering, series, directory, window=None, psi=None, clfs=None, patternlen=None):
directory = Path(directory)
prop_cycle = plt.rcParams['axes.prop_cycle']
colors = prop_cycle.by_key()['color']
cluster_to_generalized_super_instances = clustering.get_cluster_to_generalized_super_instance_map()
for c_idx, cluster in enumerate(clustering.clusters):
cluster_pts = set(cluster.get_all_points())
for s_idx, generalized_superinstance in enumerate(cluster_to_generalized_super_instances[cluster]):
labels = np.zeros((series.shape[0],), dtype=int)
generalized_superinstance_indices = []
for si in generalized_superinstance:
generalized_superinstance_indices.extend(si.indices)
labels[generalized_superinstance_indices] = 1
ignore_idxs = cluster_pts - set(generalized_superinstance_indices)
# TODO: plot all generalized superinstance representatives?
weights, importances = dtww.compute_weights_using_dt(
series, labels, generalized_superinstance[0].representative_idx,
window=window, min_ig=0.01, min_purity=0.8, max_clfs=clfs,
only_max=False, strict_cl=True, ignore_idxs=ignore_idxs,
warping_paths_fnc=dtw.warping_paths, psi=psi, patternlen=patternlen)
fig, ax = plt.subplots(nrows=2, ncols=1)
dtww.plot_margins(series[generalized_superinstance[0].representative_idx,:], weights, ax=ax[0],
importances=importances)
ml_cnt, cl_cnt, ig_cnt = 0, 0, 0
for serie_idx, serie in enumerate(series):
if serie_idx in ignore_idxs:
ig_cnt += 1
continue
label = int(labels[serie_idx])
if label == 0:
cl_cnt += 1
elif label == 1:
ml_cnt += 1
else:
raise Exception(f"Unknown label: {label}")
color = colors[label]
ax[1].plot(serie, '-', color=color, alpha=0.1 + label * 0.4)
ax[1].set_title(f"This SI size: {ml_cnt}, Other SI size: {cl_cnt}, Ignored SI size: {ig_cnt}")
plt.savefig(str(directory / f"cluster_margins_{c_idx}_{s_idx}.png"))
logger.debug(f"Created cluster_margins_{c_idx}_{s_idx}.png")
plt.close(fig)
def plotclustermargins(final_clustering, series, directory=None, window=None, clfs=None):
#directory = Path(directory)
final_clustering = np.array(final_clustering)
for clusterid in np.unique(final_clustering):
logger.info(f"Plotting cluster {clusterid}")
# TODO: Should be medoids
prototypeidx = np.where(final_clustering == clusterid)[0][0]
# np.savetxt(str(directory / f"labels_{clusterid}.csv"), final_clustering, delimiter=',', fmt='%i')
# np.savetxt(str(directory / f"series_{clusterid}.csv"), series, delimiter=',')
labels = np.zeros(final_clustering.shape)
labels[final_clustering == clusterid] = 1
weights, importances = dtww.compute_weights_using_dt(series, labels, prototypeidx,
window=window, min_ig=0.1,
max_clfs=clfs,
only_max=False, strict_cl=True)
fig, ax = dtww.plot_margins(series[prototypeidx], weights)