Python绘制主成分分析图(附数据和代码)

import pandas as pdimport numpy as npimport matplotlib.pyplot as pltfrom sklearn.decomposition import PCAfrom sklearn.preprocessing import StandardScalerfrom matplotlib.patches import Ellipsefrom matplotlib.gridspec import GridSpecfrom scipy.stats import gaussian_kde, chi2plt.rcParams['font.family'] = 'Times New Roman'# 文件路径CSV_FILE = "pca_data.csv"OUT_FIG = "pca_plot.png"# 数据生成核心参数RANDOM_SEED = 42N_PER_GROUP = 50NOISE_STD = 0.2# 真实载荷矩阵（核心）TRUE_LOADINGS = np.array([    [-0.9, 0.0],    [0.8, 0.1],    [0.3, 0.9],    [0.1, -0.9],    [-0.7, -0.6],    [-0.5, 0.7]])# 组配置（核心）GROUP_CONFIGS = {    "Group1": {"center": [-3, 3], "scale": [1.2, 0.8]},    "Group2": {"center": [3, -1], "scale": [0.9, 1.5]},    "Group3": {"center": [-1, -4], "scale": [1.5, 0.7]},}# 基础列名与颜色FEATURE_COLS = [f"Feature{i}" for i in range(1, 7)]GROUP_COL = "Group"COLORS = ["#b0d1a6", "#f5c5a0", "#e7a39a"]# 图形核心参数（删减边缘图、箭头等次要参数）FIGSIZE = (10, 10)GS_NROWS, GS_NCOLS = 4, 4SCATTER_S = 90SCATTER_ALPHA = 0.9ELLIPSE_ALPHA = 0.25

def generate_scattered_data(filename: str = CSV_FILE) -> str:    """生成模拟的PCA数据并保存为CSV文件"""    np.random.seed(RANDOM_SEED)    all_data = []    for group_name, cfg in GROUP_CONFIGS.items():        # 生成主成分得分        s_pc1 = np.random.normal(cfg["center"][0], cfg["scale"][0], N_PER_GROUP)        s_pc2 = np.random.normal(cfg["center"][1], cfg["scale"][1], N_PER_GROUP)        scores = np.vstack([s_pc1, s_pc2]).T        # 生成原始特征数据（删减噪声添加的注释）        x_raw = scores @ TRUE_LOADINGS.T + np.random.normal(0, NOISE_STD, (N_PER_GROUP, len(FEATURE_COLS)))        df_group = pd.DataFrame(x_raw, columns=FEATURE_COLS)        df_group[GROUP_COL] = group_name        all_data.append(df_group)    df_total = pd.concat(all_data, ignore_index=True)    df_total.loc[:, FEATURE_COLS] += 10  # 直接写死偏移量，删减FEATURE_OFFSET参数    df_total.to_csv(filename, index=False)    return filename

def plot_margin_type_5(ax_top, ax_right, group_data_list, ax_main, bins: int = 25):    x_min, x_max = ax_main.get_xlim()    y_min, y_max = ax_main.get_ylim()    x_bins = np.linspace(x_min, x_max, bins + 1)    y_bins = np.linspace(y_min, y_max, bins + 1)    x_hist_data, y_hist_data = [], []    for _, data, color, _marker in group_data_list:        x_counts, _ = np.histogram(data["PC1"], bins=x_bins)        y_counts, _ = np.histogram(data["PC2"], bins=y_bins)        x_hist_data.append((x_counts, color))        y_hist_data.append((y_counts, color))    x_centers = (x_bins[:-1] + x_bins[1:]) / 2    x_bottom = np.zeros(len(x_centers))    for counts, color in x_hist_data:        ax_top.bar(x_centers, counts, width=(x_bins[1] - x_bins[0]), bottom=x_bottom, color=color, alpha=0.35)        x_bottom += counts    y_centers = (y_bins[:-1] + y_bins[1:]) / 2    y_bottom = np.zeros(len(y_centers))    for counts, color in y_hist_data:        ax_right.barh(y_centers, counts, height=(y_bins[1] - y_bins[0]), left=y_bottom, color=color, alpha=0.35)        y_bottom += counts    ax_dens_x = ax_top.twinx()    ax_dens_y = ax_right.twiny()    x_plot = np.linspace(x_min, x_max, 200)    y_plot = np.linspace(y_min, y_max, 200)    for _, data, color, _marker in group_data_list:        x_kde = gaussian_kde(data["PC1"])        ax_dens_x.plot(x_plot, x_kde(x_plot), color=color, linewidth=2.5, alpha=0.9)        y_kde = gaussian_kde(data["PC2"])        ax_dens_y.plot(y_kde(y_plot), y_plot, color=color, linewidth=2.5, alpha=0.9)    ax_top.tick_params(axis="x", bottom=False, labelbottom=False)    ax_right.tick_params(axis="y", left=False, labelleft=False)    ax_dens_x.tick_params(axis="y", right=False, labelright=False)    ax_dens_y.tick_params(axis="x", top=False, labeltop=False)

def add_confidence_ellipse_95(ax, x: np.ndarray, y: np.ndarray, color: str):    cov = np.cov(x, y)    vals, vecs = np.linalg.eigh(cov)    order = vals.argsort()[::-1]    vals = vals[order]    vecs = vecs[:, order]    vx, vy = vecs[:, 0]    theta = np.degrees(np.atan2(vy, vx))    scale = np.sqrt(chi2.ppf(0.95, df=2))    width = 2 * scale * np.sqrt(vals[0])    height = 2 * scale * np.sqrt(vals[1])    ax.add_patch(Ellipse(xy=(np.mean(x), np.mean(y)), width=width, height=height, angle=theta, facecolor=color, alpha=ELLIPSE_ALPHA, edgecolor=color, zorder=2))def add_loading_vectors(ax, loadings: np.ndarray):    for i, feat in enumerate(FEATURE_COLS):        vx, vy = loadings[i, 0] * 5.0, loadings[i, 1] * 5.0        ax.arrow(0, 0, vx, vy, color="black", alpha=0.8, head_width=0.15, lw=1.2)        ax.text(vx * 1.1, vy * 1.1, feat, fontsize=14, fontweight="bold")

csv_file = generate_scattered_data(CSV_FILE)df = pd.read_csv(csv_file)X_scaled = StandardScaler().fit_transform(df[FEATURE_COLS])pca = PCA(n_components=2)pca_res = pca.fit_transform(X_scaled)loadings = pca.components_.Tpca_df = pd.DataFrame({"PC1": pca_res[:, 0], "PC2": pca_res[:, 1], GROUP_COL: df[GROUP_COL]})fig = plt.figure(figsize=FIGSIZE)gs = GridSpec(GS_NROWS, GS_NCOLS, hspace=0.0, wspace=0.0)ax_main = fig.add_subplot(gs[1:4, 0:3])ax_top = fig.add_subplot(gs[0, 0:3], sharex=ax_main)ax_right = fig.add_subplot(gs[1:4, 3], sharey=ax_main)groups = df[GROUP_COL].unique()plot_data_for_margin = []for i, group in enumerate(groups):    data = pca_df[pca_df[GROUP_COL] == group]    color = COLORS[i % len(COLORS)]    plot_data_for_margin.append((group, data, color, "o"))    ax_main.scatter(data["PC1"], data["PC2"], c=color, s=SCATTER_S, alpha=SCATTER_ALPHA, edgecolors="white")    add_confidence_ellipse_95(ax_main, data["PC1"].to_numpy(), data["PC2"].to_numpy(), color)    add_loading_vectors(ax_main, loadings)    ax_main.axhline(0, color="gray", ls="--", alpha=0.4)    ax_main.axvline(0, color="gray", ls="--", alpha=0.4)    plot_margin_type_5(ax_top, ax_right, plot_data_for_margin, ax_main)    plt.savefig(OUT_FIG, dpi=600, bbox_inches="tight")

🌿 今日的分享就到这里啦~如果这些内容有为你带来帮助，欢迎轻点右下角的【👍赞】和【👀在看】，也欢迎分享给更多需要的人，感恩~

数据和代码怎么获取？

点击关注后，后台回复关键词：

2026_map_008可直接获取完整的示例数据和代码

如有帮助，您的点赞、评论、转发是我持续创作的动力~