我正在使用sklearn的K-Means聚类,并希望使用训练好的K-Means模型将计算好的K-Means聚类标签替换为中心值。
我使用的代码如下。
# Initialize K-Means clustering model-
kmeans_conv1 = KMeans(n_clusters = 5)
# Train model on training data (compute k-means clustering)-
kmeans_conv1.fit(conv1_nonzero.reshape(-1, 1))
# number of clusters used-
kmeans_conv1.n_clusters
# 5
# Get centroids-
kmeans_conv1.cluster_centers_
'''
array([[-0.05669265],
[ 0.06742188],
[-0.08835593],
[ 0.03749201],
[ 0.0896403 ]], dtype=float32)
'''
# Clustered labels of each data point-
kmeans_conv1.labels_
set(kmeans_conv1.labels_)
Out[142]: {0, 1, 2, 3, 4}
# Get clustered label for each data point-
clustered_labels = kmeans_conv1.labels_
目前,我使用if -else条件 将标签映射到中心值,如:
new_clusters = []
for clabel in clustered_labels:
if clabel == 0:
new_clusters.append(kmeans_conv1.cluster_centers_[0][0])
elif clabel == 1:
new_clusters.append(kmeans_conv1.cluster_centers_[1][0])
elif clabel == 2:
new_clusters.append(kmeans_conv1.cluster_centers_[2][0])
elif clabel == 3:
new_clusters.append(kmeans_conv1.cluster_centers_[3][0])
elif clabel == 4:
new_clusters.append(kmeans_conv1.cluster_centers_[4][0])
最后,我想让 "new_clusters "列表或np.array变量包含中心值,而不是聚类标签。
然而,有没有更好的方法可以在不使用if-else条件的情况下实现这个目标?
这就够了。
for clabel in clustered_labels:
new_clusters.append(
kmeans_conv1.cluster_centers_[clabel][0]
)
找到了这个方法
# First conv layer condition-
cond_conv1 = [clustered_labels == 0, clustered_labels == 1, clustered_labels == 2, clustered_labels == 3, clustered_labels == 4]
# values-
val_conv1 = kmeans_conv1.cluster_centers_[:, 0]
# Get new clustered value weights-
new_weights_conv1 = np.select(cond_conv1, val_conv1)