金融知识图谱嵌入

WHENWhen implementing temporal data splitting for knowledge graphs

ACTIONuse time-ordered train/val/test split with train_edges occurring before val_edges and test_edges temporally

CONSEQUENCETemporal knowledge graphs will suffer from look-ahead bias where model sees future events during training, causing inflated metrics that do not reflect real-world temporal prediction performance

WHENWhen storing temporal edge data in DGLGraph edata

ACTIONconvert time values to float32 dtype to verify numerical precision for time interval calculations

CONSEQUENCETime interval calculations (G.time_interval = min(time_diff)) will produce incorrect or inconsistent results due to dtype mismatches, breaking the time interval transform pipeline in DKG/model/time_interval_transform.py

WHENWhen storing node type identifiers in DGLGraph ndata

ACTIONconvert node type values to integer type matching the graph idtype to verify consistent graph operations

CONSEQUENCEGraph operations (edge_subgraph, degree calculations) will fail or produce incorrect results due to dtype mismatch between node indices and node type identifiers