ax.plot(inner_ring[0, i:i + 2], inner_ring[1, i:i + 2])
ax.plot(outer_ring[0, i:i + 2], outer_ring[1, i:i + 2])
plt.xlim(xmin, xmax)
plt.ylim(ymin, ymax)
cur_axes = plt.gca()
cur_axes.axes.get_xaxis().set_ticks([])
cur_axes.axes.get_yaxis().set_ticks([])
if SAVE_SVG:
plt.savefig("euclidean_center.svg", bbox_inches='tight')
if SAVE_PNG:
plt.savefig("euclidean_center.png", dpi=300, bbox_inches='tight')
all_rings = [ring.T for ring in all_rings]
vwb = VOT(all_rings[0][0:-1:5, :], all_rings, verbose=False)
tick = time.clock()
vwb.cluster(lr=0.5, max_iter_h=3000, max_iter_y=1)
tock = time.clock()
print(tock - tick)
if SAVE_OR_PLOT_FIGURE:
fig = plt.figure(figure_id, figsize=(2, 2))
ax = fig.add_subplot(111)
for i in range(N):
plt.plot(vwb.x[i][:, 0], vwb.x[i][:, 1], linewidth=5, c='lightgray')
size = vwb.y.shape[0] // 2
ax.set_prop_cycle(
cycler(color=[cm(1. * i / (size - 1)) for i in range(size - 1)]))
from sklearn.cluster import KMeans
kmeans = KMeans(n_clusters=K, init=y).fit(x)
label = kmeans.predict(x)
y = kmeans.cluster_centers_
color_map = np.array([[237, 125, 49, 255], [112, 173, 71, 255], [91, 155, 213, 255]]) / 255
# ---------------VWB---------------
for reg in [0.5, 2, 1e9]:
y_copy = y.copy()
x_copy = x.copy()
vot = VOT(y_copy, [x_copy], verbose=False)
vot.cluster(lr=0.5, max_iter_h=1000, max_iter_y=1, beta=0.5, reg=reg)
idx = vot.idx
fig = plt.figure(figsize=(4, 4))
ce = color_map[idx[0]]
utils.scatter_otsamples(vot.y, vot.x[0], size_p=30, marker_p='o', color_x=ce,
xmin=xmin, xmax=xmax, ymin=ymin, ymax=ymax, facecolor_p='none')
plt.axis('off')
# plt.savefig(str(int(reg)) + ".svg", bbox_inches='tight')
plt.savefig(str(reg) + ".png", dpi=300, bbox_inches='tight')
plt.figure(figsize=(4, 4))
cov1 = [[0.03, 0], [0, 0.01]]
x1, y1 = np.random.multivariate_normal(mean1, cov1, 5000).T
x1 = np.stack((x1, y1), axis=1).clip(-0.99, 0.99)
mean2 = [0.5, 0.5]
cov2 = [[0.01, 0.], [0., 0.03]]
x2, y2 = np.random.multivariate_normal(mean2, cov2, 1000).T
x2 = np.stack((x2, y2), axis=1).clip(-0.99, 0.99)
mean = [0.0, -0.5]
cov = [[0.02, 0], [0, 0.02]]
K = 50
x, y = np.random.multivariate_normal(mean, cov, K).T
x = np.stack((x, y), axis=1).clip(-0.99, 0.99)
vot = VOT(x, [x1, x2], verbose=False)
vot.cluster(max_iter_h=3000, max_iter_y=1)
idx = vot.idx
xmin, xmax, ymin, ymax = -1., 1., 0., 1.
for k in [21]:
plt.figure(figsize=(8, 4))
for i in range(2):
ce = np.array(plt.get_cmap('viridis')(idx[i] / (K - 1)))
utils.scatter_otsamples(vot.y,
vot.x[i],
size_p=30,
marker_p='o',
color_e=ce,
xmin=xmin,
cys = cys_base[labels]
ys, xs = 15, 3
plt.figure(figsize=(12, 8))
plt.subplot(231)
plt.xlim(xmin, xmax)
plt.ylim(ymin, ymax)
plt.grid(True)
plt.title('w/o reg before')
plt.scatter(x[:, 0], x[:, 1], s=xs, color=utils.COLOR_LIGHT_GREY)
for p, cy in zip(y, cys):
plt.scatter(p[0], p[1], s=ys, color=cy)
# ------- run WM -------- #
vot = VOT(y.copy(), [x.copy()], label_y=labels, verbose=False)
print("running Wasserstein clustering...")
tick = time.time()
vot.cluster(max_iter_y=1)
tock = time.time()
print("total running time : {0:g} seconds".format(tock - tick))
cxs = cxs_base[vot.label_x[0]]
# ------ plot map ------- #
fig232 = plt.subplot(232)
plt.xlim(xmin, xmax)
plt.ylim(ymin, ymax)
plt.grid(True)
plt.title('w/o reg map')
for p, p0 in zip(vot.y, vot.y_original):
downsample = 100 x1_down = x1[0:-1:downsample, :] # ------- run WM -------- # iterP = 8 downsample = 10 x = x1_down.copy() x1_copy = x1.copy() x2_copy = x2.copy() x3_copy = x3.copy() vwb = VOT(x, [x1_copy, x2_copy, x3_copy], verbose=False) output = vwb.cluster(lr=1, max_iter_h=3000, max_iter_y=iterP, lr_decay=500, beta=0.5, icp=True) fig2 = plt.figure(figsize=(8, 8)) # ax2 = fig2.add_subplot(111, projection='3d') # outE1 = vwb.x[0] outE2 = vwb.x[1] outE3 = vwb.x[2] outP = vwb.y ax2.scatter(x1[:, 0], x1[:, 1], x1[:, 2], s=dot_size, color=color_map[2], alpha=alpha) ax2.scatter(x2[:, 0], x2[:, 1], x2[:, 2], s=dot_size, color=color_map[1], alpha=alpha) ax2.scatter(x3[:, 0], x3[:, 1], x3[:, 2], s=dot_size, color=color_map[2], alpha=alpha)
y1 = np.cos(u0) * np.sin(v0)
y2 = np.sin(u0) * np.sin(v0)
y3 = np.cos(v0)
plt.show()
x1 = np.stack((x11, x12, x13), axis=1).clip(-0.99, 0.99)
x2 = np.stack((x21, x22, x23), axis=1).clip(-0.99, 0.99)
y = np.stack((y1, y2, y3), axis=1).clip(-0.99, 0.99)
y /= np.linalg.norm(y, axis=1, keepdims=True)
x1 /= np.linalg.norm(x1, axis=1, keepdims=True)
x2 /= np.linalg.norm(x2, axis=1, keepdims=True)
vwb = VOT(y, [x1, x2], verbose=False)
output = vwb.cluster(max_iter_h=5000, max_iter_y=1, space='spherical')
idx = output['idx']
xmin, xmax, ymin, ymax = -1.0, 1.0, -0.5, 0.5
u, v = np.mgrid[np.pi/4:np.pi*5/4:1000j, np.pi/2:np.pi*3/2:1000j]
gx = np.cos(u)*np.sin(v)
gy = np.sin(u)*np.sin(v)
gz = np.cos(v)
for k in [12]:
fig = plt.figure(figsize=(8, 8))
ax = fig.add_subplot(111, projection='3d')
colors = plt.cm.magma((gx - gx.min()) / float((gx - gx.min()).max()))
ax.plot_surface(gx * 0.95, gy * 0.95, gz * 0.95, antialiased=False, facecolors=colors, linewidth=0, shade=False)
for i in range(2):
cov1 = [[0.02, 0], [0, 0.02]]
x1, y1 = np.random.multivariate_normal(mean1, cov1, 5000).T
x1 = np.stack((x1, y1), axis=1).clip(-0.99, 0.99)
mean2 = [0.5, 0.]
cov2 = [[0.02, 0], [0, 0.02]]
x2, y2 = np.random.multivariate_normal(mean2, cov2, 5000).T
x2 = np.stack((x2, y2), axis=1).clip(-0.99, 0.99)
mean = [0.0, -0.5]
cov = [[0.02, 0], [0, 0.02]]
K = 50
x, y = np.random.multivariate_normal(mean, cov, K).T
x = np.stack((x, y), axis=1).clip(-0.99, 0.99)
vot = VOT(x, [x1, x2], verbose=False)
output = vot.cluster(max_iter_h=5000, max_iter_y=1)
idx = vot.idx
xmin, xmax, ymin, ymax = -1.0, 1.0, -0.5, 0.5
for k in [33]:
plt.figure(figsize=(8, 4))
for i in range(2):
ce = np.array(plt.get_cmap('viridis')(idx[i] / (K - 1)))
utils.scatter_otsamples(vot.y,
vot.x[i],
size_p=30,
marker_p='o',
color_x=ce,
xmin=xmin,
cys = cys_base[labels]
ys, xs = 15, 3
plt.figure(figsize=(12, 7))
plt.subplot(231)
plt.xlim(xmin, xmax)
plt.ylim(ymin, ymax)
plt.grid(True)
plt.title('w/o reg before')
plt.scatter(x[:, 0], x[:, 1], s=xs, color=utils.COLOR_LIGHT_GREY)
for p, cy in zip(y, cys):
plt.scatter(p[0], p[1], s=ys, color=cy)
# ------- run WM -------- #
vot = VOT(y.copy(), [x.copy()], label_y=labels, verbose=False)
print("running Wasserstein clustering...")
tick = time.time()
vot.cluster(lr=0.5, max_iter_y=1)
tock = time.time()
print("total running time : {0:.4f} seconds".format(tock - tick))
cxs = cxs_base[vot.label_x[0]]
# ------ plot map ------- #
fig232 = plt.subplot(232)
plt.xlim(xmin, xmax)
plt.ylim(ymin, ymax)
plt.grid(True)
plt.title('w/o reg map')
for p, p0 in zip(vot.y, vot.y_original):
mean2 = [0.5, 0.]
cov2 = [[0.02, 0], [0, 0.02]]
x2, y2 = np.random.multivariate_normal(mean2, cov2, 5000).T
x2 = np.stack((x2, y2), axis=1).clip(-0.99, 0.99)
wd = np.zeros((50, 50))
k = 0
for K in [50, 125, 250, 500, 750, 1000, 1500, 2500]:
for i in range(10):
mean = [0.0, -0.5]
cov = [[0.02, 0], [0, 0.02]]
x, y = np.random.multivariate_normal(mean, cov, K).T
x = np.stack((x, y), axis=1).clip(-0.99, 0.99)
vwb = VOT(x, [x1, x2], verbose=False)
output = vwb.cluster(lr=1, max_iter_h=10000, max_iter_y=1, beta=0.9, lr_decay=500)
wd[k, i] = output['wd']
print(output['wd'])
# np.savetxt(f, output['wd'], delimiter=',')
del x
del vwb
k += 1
# np.savetxt('ship_error/K_total.csv', wd, delimiter=',')
import ot
import ot.plot
M = ot.dist(x1, x2, 'sqeuclidean')
plt.figure(figsize=(8, 8))
minx, maxx, miny, maxy = -1, 1, -1, 1
plot_map = True
# -------------------- #
# ------- VOT -------- #
# -------------------- #
print('---- Running VOT ----')
dist = cdist(y, x, 'sqeuclidean')
mass_e = np.ones(N0) / N0
mass_p = np.ones(K) / K
tick = time.clock()
vot = VOT(y=y, x=x, verbose=False)
output = vot.cluster(max_iter_y=1, max_iter_h=3000, lr=1, lr_decay=200, beta=0.9)
tock = time.clock()
print('total time: {0:.4f}'.format(tock-tick))
if plot_map:
# ----- plot before ----- #
plt.subplot(331)
plt.xlim(minx, maxx)
plt.ylim(miny, maxy)
plt.grid(True)
plt.title('before')
plt.scatter(vot.x[0][:, 0], vot.x[0][:, 1], marker='x', color=utils.COLOR_RED)
plt.scatter(p_coor_before[:, 0], p_coor_before[:, 1], marker='+', color=utils.COLOR_DARK_BLUE)
marker_p='o',
color_e=ce,
xmin=xmin,
xmax=xmax,
ymin=ymin,
ymax=ymax,
facecolor_p='none')
plt.axis('off')
# plt.savefig("kmeans.svg", bbox_inches='tight')
plt.savefig("kmeans.png", dpi=300, bbox_inches='tight')
# --------------- OT ---------------
y_copy = y.copy()
x_copy = x.copy()
vwb = VOT(y_copy, [x_copy], verbose=False)
output = vwb.cluster(lr=0.5,
max_iter_h=20,
max_iter_y=1,
beta=0.5,
keep_idx=True)
idx, idxs = output['idx'], output['idxs'][0]
for i in range(0, min(21, len(idxs))):
fig = plt.figure(figsize=(4, 4))
ce = color_map[idxs[i]]
utils.scatter_otsamples(vwb.y,
vwb.x[0],
nop=True,
size_p=30,
marker_p='o',
color='k',
zorder=1)
ax6.set_xlabel('R')
ax6.set_ylabel('G')
ax6.set_zlabel('B')
# fig6.savefig("x3_histogram_kmeans1.svg", bbox_inches='tight')
fig6.savefig("x3_histogram_kmeans1.png", dpi=300, bbox_inches='tight')
# -------------- VWB ------------------ #
x1 = x1.clip(-0.99, 0.99)
x2 = x2.clip(-0.99, 0.99)
x3 = x3.clip(-0.99, 0.99)
x = C1_16.clip(-0.99, 0.99)
vwb = VOT(x, [x1], verbose=False)
vwb.cluster(lr=1, max_iter_h=5000, max_iter_y=1)
idx = vwb.idx[0]
x1_vwb_16 = vwb.y[idx, :]
x1_vwb_16 = np.transpose(np.reshape(x1_vwb_16 * 255, (128, 128, 3)), (1, 0, 2))
imageio.imwrite("x1_vwb_16.png", x1_vwb_16)
fig1 = plt.figure(figsize=(8, 8))
ax1 = fig1.add_subplot(111, projection='3d')
ce2 = vwb.y[idx]
p11 = vwb.y
ax1.scatter(x1[:, 0], x1[:, 1], x1[:, 2], s=dot_size, color=ce2, alpha=alpha)
ax1.xaxis.pane.fill = False
ax1.yaxis.pane.fill = False
ax1.zaxis.pane.fill = False
ax1.scatter(p11[:, 0],
p11[:, 1],