Umap和t-SNE降维可视化 | 降维 Ⅴ | SLie's Blog|琴弦之轮

t-SNE

t-SNE 可视化降维方法简介 - Pand`s Blog (pandint.github.io)
详解可视化利器 t-SNE 算法：数无形时少直觉 | 机器之心 (jiqizhixin.com)

from sklearn.cluster import SpectralClustering
from sklearn.manifold import TSNE
import seaborn as sns

tsne = TSNE(
    n_components=2,
    perplexity=30,   # 控制局部和全局结构之间的平衡，默认值通常在5到50之间
    learning_rate='auto',
    init='pca',
    random_state=1024   # 设置随机种子确保结果可复现
)

# 运行t-SNE算法进行降维可视化
embedded_data = tsne.fit_transform(X_reduced)

fig, axs = plt.subplots(1,4, sharex=True, sharey=True, figsize=(15,4))

for i, ax in enumerate(axs.flat):
    spec = SpectralClustering(n_clusters=i+3)
    spec.fit(X_reduced)

    # kmeans = KMeans(n_clusters=i+3, n_init='auto')
    # kmeans.fit(X_reduced)

    # sns.set_context({'figure.figsize':[10, 10]})
    palette = sns.color_palette("bright", i+3)
    sns.scatterplot(x=embedded_data.T[0], y=embedded_data.T[1], hue=spec.labels_ ,palette=palette, ax=ax)     
    ax.set_title(f"n_clusters={i+3}")
    # print(i, calinski_harabasz_score(X_reduced, spec.labels_))

plt.show()