def visualizeEmbedding_2D_withCluster(ncluster=3):
twoGraphEmbeds, twoGRids = retrieveCrossIntervalEmbeddings(
"../miscs/taxi-deepwalk-CA-usespatial.vec", skipheader=0)
gndTid, gndLabels = generate_AC_clusteringLabel(ncluster, "rac")
# gndTid, gndLabels = generatePOIClusteringlabel(ncluster)
# gndTid, gndLabels = generate_od_clusteringLabel(ncluster)
clrs = ["b", "r", "g", "w", "c", "b"]
print[len(np.argwhere(gndLabels == i)) for i in range(ncluster)]
plt.figure(figsize=(40, 36))
plt.suptitle("RAC count as ground truth {0} clusters".format(ncluster))
for h in range(len(twoGRids)):
plt.subplot(4, 6, h + 1)
for cluster in range(ncluster):
groupIds = gndTid[np.argwhere(gndLabels == cluster)]
idx = np.in1d(twoGRids[h], groupIds)
x = twoGraphEmbeds[h][idx, 0]
y = twoGraphEmbeds[h][idx, 1]
ids = twoGRids[h][idx]
plt.scatter(x, y, c=clrs[cluster], hold=True)
plt.xlim([-1.0, 1])
plt.ylim([-1, 0.5])
for i, e in enumerate(ids):
plt.annotate(s=str(e),
xy=(x[i], y[i]),
xytext=(-5, 5),
textcoords="offset points")
plt.title("2D visualization at {0}".format(h))
plt.savefig("CA-RAC-{0}cluster.png".format(ncluster))
def getDeepwalkEmbeddingFeatures(Spatial):
res = {}
emptyEmbedding = np.zeros((8,))
features, ids = retrieveCrossIntervalEmbeddings("../miscs/{0}/taxi-deepwalk-CA-{1}.vec".format(year, Spatial), skipheader=0)
for h in range(24):
sids = np.argsort(ids[h])
res[h] = features[h][sids, :]
if len(sids) != 77:
print len(sids), h
for i in range(1, 78):
if i not in ids[h]:
res[h] = np.insert(res[h], i-1, emptyEmbedding, axis=0)
assert res[h].shape == (77, 8)
return res
def getDeepwalkEmbeddingFeatures(Spatial):
with open("../miscs/POI_tract.pickle") as fin:
ordKey = pickle.load(fin)
tract_poi = pickle.load(fin)
res = {}
emptyEmbedding = np.zeros((20, ))
features, ids = retrieveCrossIntervalEmbeddings(
"../miscs/{0}/taxi-deepwalk-tract-{1}.vec".format(year, Spatial),
skipheader=0)
for h in range(8):
sids = np.argsort(ids[h])
res[h] = features[h][sids, :]
if len(sids) != 801:
print len(sids), h
for i, k in enumerate(ordKey):
if k not in ids[h]:
res[h] = np.insert(res[h], i, emptyEmbedding, axis=0)
assert res[h].shape == (801, 20)
return res
def visualizeEmbedding_2D():
twoGraphEmbeds, twoGRids = retrieveCrossIntervalEmbeddings(
"../miscs/taxi-deepwalk-CA-usespatial-2D.vec", skipheader=0)
groups = [[13, 14, 15, 16], [8, 32, 33], [44, 45, 47, 48], [76]]
# groups = [[5,6,7,21,22], [8,32,33], [26,27,29,30]]
clrs = ["b", "r", "g", "c", "w", "k"]
plt.figure(figsize=(12, 6))
for k, h in enumerate([7, 8, 9, 16, 17, 18, 21,
22]): # enumerate(range(len(twoGRids))):
f = plt.subplot(2, 4, k + 1)
xr = 0
yr = 0
xo = 0
yo = 0
for i, group in enumerate(groups):
idx = np.in1d(twoGRids[h], group)
x = twoGraphEmbeds[h][idx, 0]
y = twoGraphEmbeds[h][idx, 1]
ids = twoGRids[h][idx]
plt.scatter(x, y, s=40, c=clrs[i], hold=True)
# plt.xlim([-3.0, 0.6])
# plt.ylim([-2, 0.5])
for j, e in enumerate(ids):
if e in [76, 47]:
plt.annotate(s=str(e),
xy=(x[j], y[j]),
xytext=(-2, -15),
textcoords="offset points",
fontsize=14)
elif e in [14, 15, 16]:
xr += x[j]
yr += y[j]
elif e in [8, 32, 33]:
xo += x[j]
yo += y[j]
else:
if k == 0:
if e == 13:
plt.annotate(s=str(e),
xy=(x[j], y[j]),
xytext=(-2, -15),
textcoords="offset points",
fontsize=14)
else:
plt.annotate(s=str(e),
xy=(x[j], y[j]),
xytext=(0, -15),
textcoords="offset points",
fontsize=14)
if k == 0:
plt.annotate(s="14,15,16",
xy=(xr / 3, yr / 3),
xytext=(5, 3),
textcoords="offset points",
fontsize=14)
plt.annotate(s="8,32,33",
xy=(xo / 3, yo / 3),
xytext=(-50, 8),
textcoords="offset points",
fontsize=14)
plt.title("{0}:00".format(h), fontsize=16)
f.tick_params(axis="both", labelsize=10)
plt.tight_layout()
plt.savefig("CA-case-3region.pdf")