如何生成带黑边的圆形LDA主题词云并应用于PCA图表

需求说明 #

适配出版规范，为LDA生成的每个主题制作圆形、带黑色边框的词云图，最终可将生成的词云应用到PCA图表中。

核心修改点 #

• 新增圆形掩膜参数限制词云生成范围，实现圆形效果
• 子图渲染时添加与词云同尺寸的圆形黑色边框，符合出版规范
• 保留词云对象输出逻辑，支持后续嵌入PCA图表

完整可运行代码 #

import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
import numpy as np
from wordcloud import WordCloud
sns.set()

# 输入数据
topics_df = pd.DataFrame({
    'Unnamed: 0': {0: 'Topic1', 1: 'Topic2', 2: 'Topic3', 3: 'Topic4', 4: 'Topic5'}, 
    'Terms per Topic': {
        0: 'charge, model, infrastructure, station, time, base, propose, drive, public, service, range, problem, study, paper, network, user, travel, battery, city, method', 
        1: 'cost, emission, fuel, car, high, price, low, reduce, hybrid, carbon, battery, gas, alternative, benefit, compare, find, conventional, efficiency, gasoline, total', 
        2: 'market, technology, policy, paper, change, development, support, mobility, business, industry, government, focus, study, transition, innovation, decision, develop, technological, case, lead', 
        3: 'consumer, policy, adoption, effect, model, purchase, bev, factor, evs, study, incentive, choice, preference, subsidy, influence, environmental, state, level, suggest, analysis', 
        4: 'energy, system, electricity, demand, transport, scenario, impact, increase, power, phev, transportation, sector, potential, base, show, consumption, reserve, term, economic, grid'
    }
})

# 生成圆形掩膜：400*400像素正方形，中间圆形区域为可生成词云的区域
x, y = np.ogrid[:400, :400]
mask = (x - 200) ** 2 + (y - 200) ** 2 > 190 ** 2
mask = 255 * mask.astype(int)

# 词云初始化新增mask参数
wc = WordCloud(
    background_color="white", 
    colormap="Dark2",
    max_font_size=150, 
    random_state=42,
    mask=mask,
    contour_width=0 # 禁用自带边框，自定义绘制更灵活
)

plt.rcParams['figure.figsize'] = [20, 15]
# 存储所有主题词云对象，后续嵌入PCA时可直接调用
topic_wc_list = []

# 创建每个主题的子图
for i in range(5):
    wc.generate(text=topics_df["Terms per Topic"][i])
    topic_wc_list.append(wc.to_array()) # 保存词云数组
    plt.subplot(5, 4, i+1)
    plt.imshow(wc, interpolation="bilinear")
    # 绘制黑色圆形边框，和词云大小匹配
    ax = plt.gca()
    circle = plt.Circle((200, 200), 190, color='black', fill=False, linewidth=2)
    ax.add_patch(circle)
    plt.axis("off")
    plt.title(topics_df["Unnamed: 0"][i], fontsize=12)
plt.show()

PCA嵌入参考逻辑 #

如果需要将生成的词云放到PCA图表对应位置，可参考以下步骤实现：

1. 提前计算所有主题的PCA降维后坐标，存储为二维数组topic_pca_coords
2. 导入相关工具类：

from matplotlib.offsetbox import OffsetImage, AnnotationBbox

3. 在PCA图表生成后，遍历坐标和词云数组，绑定到对应位置：

# pca_ax为PCA图表的坐标轴对象
for idx, (x_coord, y_coord) in enumerate(topic_pca_coords):
    img = OffsetImage(topic_wc_list[idx], zoom=0.2) # 可调整zoom参数控制词云大小
    ab = AnnotationBbox(img, (x_coord, y_coord), frameon=False)
    pca_ax.add_artist(ab)

内容的提问来源于stack exchange，提问作者Bloxx