Replication Package: Predicting Urban Economic Resilience via Graph Convolutional Networks and Functional Urban Networks (N=93 China)

Description: This repository contains the replication package for the manuscript titled "[Insert Your Paper Title]". This study investigates the transmission of economic resilience across functional urban networks using a Graph Convolutional Network (GCN) framework. Key Contributions: Sample Expansion: Addresses small-sample bias by pooling 93 prefecture-level cities across four major Chinese urban agglomerations (JJJ, YRD, PRD, CD-CQ) into a 93×93 Block-Diagonal adjacency matrix. Causal-Logic Alignment: Implements a strict temporal cutoff, utilizing pre-pandemic features (2019) to predict pandemic-impact resilience (2020-2023), mitigating information leakage. Methodological Rigor: Compares GCN performance against established machine learning baselines (Random Forest, XGBoost) and topological null models (Random Graph) using a leave-one-city-out cross-validation (LOOCV) framework. Resilience Regimes: Demonstrates the superiority of GCN in classification tasks (High/Medium/Low resilience), revealing how economic flow networks capture resilience transmission better than geographical proximity. Contents: data/: Processed feature matrices (X_pooled.npy), OD flow matrices (A_pooled.npy), and GDP-based Resilience Index (y_pooled.npy). scripts/: Standardized Python scripts for feature generation, experiment execution, and figure plotting. results/: CSV files containing the aggregated classification performance metrics. Usage: Please ensure Python 3.10+ is installed with the dependencies listed in requirements.txt. Run python run_pooled_classification.py to replicate the main classification results.描述：本仓库为题为「[请插入论文标题]」的手稿提供复现包。本研究基于图卷积网络（Graph Convolutional Network, GCN）框架，探究功能性城市网络中的经济韧性传导规律。 核心贡献： 样本拓展：通过整合中国四大国家级城市群（京津冀（JJJ, Jing-Jin-Ji）、长三角（YRD, Yangtze River Delta）、珠三角（PRD, Pearl River Delta）、成渝（CD-CQ, Chengdu-Chongqing））的93个地级市数据，构建93×93的分块对角邻接矩阵，以缓解小样本偏误。 因果逻辑对齐：采用严格的时间截断设计，以疫情前的2019年特征数据预测疫情冲击下2020-2023年的经济韧性，有效规避信息泄露风险。 方法论严谨性：通过留一城市交叉验证（Leave-One-City-Out Cross-Validation, LOOCV）框架，将GCN的表现与经典机器学习基线模型（随机森林（Random Forest）、XGBoost）以及拓扑零模型（随机图（Random Graph））进行对比。 韧性分类机制：验证了GCN在经济韧性三分分类（高/中/低韧性）任务中的优越性，揭示了经济流动网络相较于地理邻近性，更能有效捕捉韧性传导规律。 文件内容： data/：包含预处理后的特征矩阵（X_pooled.npy）、出行OD流矩阵（A_pooled.npy）以及基于GDP的韧性指数（y_pooled.npy）。 scripts/：包含特征生成、实验运行与图表绘制的标准化Python脚本。 results/：包含分类性能指标汇总的CSV文件。 使用说明：请确保已安装Python 3.10及以上版本，并安装requirements.txt中列出的依赖包。运行`python run_pooled_classification.py`即可复现核心分类实验结果。