def gap_stat_mine(data, cluster_nums, ref_num=10, cluster_method=None):
from commontool.algorithm.tool import gap_statistic
labels_list, Wks, Wks_refs_log_mean, gaps, s, k_selected = \
gap_statistic(data, cluster_nums, ref_num, cluster_method)
x = np.arange(len(cluster_nums))
cluster_nums = np.array(cluster_nums)
plt.figure()
plt.plot(x, Wks, 'b.-')
plt.xlabel('#subgroup')
plt.ylabel(('W\u2096'))
if len(x) > 20:
middle_idx = int(len(x) / 2)
plt.xticks(x[[0, middle_idx, -1]], cluster_nums[[0, middle_idx, -1]])
else:
plt.xticks(x, cluster_nums)
Wks_log = np.log(Wks)
plt.figure()
plt.plot(x, Wks_log, 'b.-')
plt.plot(x, Wks_refs_log_mean, 'r.-')
plt.xlabel('#subgroup')
plt.ylabel('log(W\u2096)')
if len(x) > 20:
middle_idx = int(len(x) / 2)
plt.xticks(x[[0, middle_idx, -1]], cluster_nums[[0, middle_idx, -1]])
else:
plt.xticks(x, cluster_nums)
return labels_list, gaps, s, k_selected
def test_gap_statistic1():
from commontool.algorithm.tool import gap_statistic, uniform_box_sampling
from matplotlib.patches import Rectangle
cluster1_x = np.random.normal(2, 0.2, 100)
cluster1_y = np.random.normal(1, 0.2, 100)
cluster1 = np.c_[cluster1_x, cluster1_y]
cluster2_x = np.random.normal(3, 0.2, 100)
cluster2_y = np.random.normal(3, 0.2, 100)
cluster2 = np.c_[cluster2_x, cluster2_y]
cluster3_x = np.random.normal(5, 0.2, 100)
cluster3_y = np.random.normal(1, 0.2, 100)
cluster3 = np.c_[cluster3_x, cluster3_y]
data = np.r_[cluster1, cluster2, cluster3]
cluster_nums = range(1, 11)
labels_list, Wks, Wks_refs_log_mean, gaps, s, k_selected = gap_statistic(data, cluster_nums)
minimums = np.atleast_2d(np.min(data, 0))
maximums = np.atleast_2d(np.max(data, 0))
bounding_box = np.r_[minimums, maximums]
dists = bounding_box[1] - bounding_box[0]
rect = Rectangle(bounding_box[0], dists[0], dists[1], edgecolor='r', facecolor='none')
uniform_samples = uniform_box_sampling(data.shape[0], bounding_box)
fig1, ax1 = plt.subplots()
ax1.set_xlim([0, 6])
ax1.set_ylim([0, 5])
ax1.scatter(data[:, 0], data[:, 1])
ax1.scatter(uniform_samples[:, 0], uniform_samples[:, 1], c='r')
ax1.add_patch(rect)
fig1.tight_layout()
x = np.arange(len(cluster_nums))
plt.figure()
plt.plot(x, Wks, 'b.-')
plt.xlabel('#cluster')
plt.ylabel(('W\u2096'))
plt.xticks(x, cluster_nums)
Wks_log = np.log(Wks)
plt.figure()
plt.plot(x, Wks_log, 'b.-')
plt.plot(x, Wks_refs_log_mean, 'r.-')
plt.xlabel('#cluster')
plt.ylabel('log(W\u2096)')
plt.xticks(x, cluster_nums)
plt.figure()
plt.plot(x, gaps, 'b.-')
plt.fill_between(x, gaps-s, gaps+s, alpha=0.5)
plt.xlabel('#cluster')
plt.ylabel('gap statistic')
plt.xticks(x, cluster_nums)
plt.show()
def test_gap_statistic2():
from commontool.algorithm.tool import gap_statistic
from commontool.algorithm.cluster import hac_scipy
cluster1_x = np.random.normal(2, 0.2, 100)
cluster1_y = np.random.normal(1, 0.2, 100)
cluster1 = np.c_[cluster1_x, cluster1_y]
cluster2_x = np.random.normal(3, 0.2, 100)
cluster2_y = np.random.normal(3, 0.2, 100)
cluster2 = np.c_[cluster2_x, cluster2_y]
cluster3_x = np.random.normal(5, 0.2, 100)
cluster3_y = np.random.normal(1, 0.2, 100)
cluster3 = np.c_[cluster3_x, cluster3_y]
data = np.r_[cluster1, cluster2, cluster3]
cluster_nums = range(1, 11)
labels_list, y, Wks_refs_log_mean, gaps, s, k_selected = gap_statistic(
data, cluster_nums, cluster_method=hac_scipy)
fig, ax = plt.subplots()
ax.set_xlim([0, 6])
ax.set_ylim([0, 5])
ax.scatter(data[:, 0], data[:, 1], c='k')
fig.tight_layout()
x = np.arange(len(cluster_nums))
plt.figure()
plt.plot(x, y, 'k.-')
plt.xlabel('k')
plt.ylabel('W\u2096')
plt.xticks(x, cluster_nums)
plt.tight_layout()
x1 = x[:-1]
y1 = [y[i] - y[i + 1] for i in x1]
fig1, ax1 = plt.subplots()
ax1.plot(x1, y1, 'k.-')
ax1.set_xlabel('k')
ax1.set_ylabel('V\u2096')
ax1.set_xticks(x1)
ax1.set_xticklabels(cluster_nums[1:])
fig1.tight_layout()
x2 = x1[:-1]
y2 = [y1[i] - y1[i + 1] for i in x2]
fig2, ax2 = plt.subplots()
ax2.plot(x2, y2, 'k.-')
ax2.set_xlabel('k')
ax2.set_ylabel('-\u25b3V\u2096')
ax2.set_xticks(x2)
ax2.set_xticklabels(cluster_nums[1:-1])
fig2.tight_layout()
plt.show()