Insights

A practical guide to mini-batch graph training in TGraphX using NeighborLoader, GraphSAINT, and ClusterLoader — the three sampling strategies that scale GNNs beyond full-batch training.

Why schema awareness matters in deep learning over structured data, and the design direction that schema-aware architectures take.

A practical introduction to TGraphX's graph reinforcement learning subsystem — environments, algorithms, and what to expect from a research-grade module labeled Experimental.

A decision guide: when flat node feature vectors work well, when they break, and how to tell which category your task falls into.

GNN benchmark numbers often look more impressive than they are. This note discusses the common evaluation shortcuts that inflate results and how to read benchmark claims more carefully.

A practical tutorial on TGraphX's knowledge graph subsystem: TransE, DistMult, ComplEx, and RotatE, with optional tensor-valued entity features.

AI coding tools produce better graph code when the API is explicit, the error messages are actionable, and the canonical surface is small. TGraphX is designed with this in mind.

Most GNN bugs are shape bugs in disguise. TGraphX provides validation utilities that catch them early, before they show up as obscure runtime errors after hours of training.

A decision-oriented comparison aimed at researchers choosing a graph learning framework for a specific project. When PyG is right, when TGraphX adds value, and how to test the fit quickly.

Reproducing published GNN results is harder than it should be. This note walks through the hidden sources of non-determinism and how explicit tooling helps.

A step-by-step deeper tutorial: validate a tensor graph, build a custom training loop with NeighborLoader, save artifacts, and reproduce results.

Most GNN frameworks assume every node is a flat vector. That assumption breaks for image patches, volumetric blocks, and sequences. Here is why node feature shape matters and how TGraphX handles it.

TGraphX is a tensor-native graph learning framework for PyTorch. This introduction explains what a TGX graph is, why tensor-valued node features matter, and where the framework fits in the GNN ecosystem.

A step-by-step tutorial for representing graph data with tensor-valued node features in TGraphX — including validation, a first training run, and common mistakes to avoid.

A balanced comparison of TGraphX and PyTorch Geometric. Both run on PyTorch, but they serve different use cases. This article explains when each framework fits, without pretending one replaces the other.