~/talk

Stop Engineering Features. Start Learning Graphs.

Talkat WhatTheStack 2026 · 2026-09-14..09-19· [machine-learning, data-engineering]
links:none on file
slides:none on file

Relational tables become a heterogeneous graph using foreign keys as edges and many‑to‑many tables as weighted edges, then a GNN learns order embeddings to predict late deliveries. On a Brazilian Kaggle e‑commerce dataset the graph model outperforms a traditional pipeline that uses 21 handcrafted features, despite slower training (1s vs 38s). Graph construction and per‑row transforms replace multi‑table joins; implementation uses PyTorch Geometric and yields higher F1/AU with minimal manual feature engineering.

AI-written summary · about

  1. 0:00Introduction and motivation
  2. 3:57From tabular data to graphs
  3. 7:56Graph types and GNN basics
  4. 17:33Relational deep learning case study
  5. 26:34Benchmark results and tradeoffs
  6. 29:43How to try this yourself
  7. 31:32Summary and takeaways
~/search
type to search 6,518 conferences, 143,918 speakers and 192,133 talks
~/help
cmdctrl k
search every conference, speaker and series
/
focus the filter search
h
home
c
conferences
s
speakers
e
series
t
light / dark — the status bar button does the same
← →
on a talk: 5 seconds back / forward
0–9
on a talk: jump to 0%–90% of the recording
?
this help
esc
close this, then clear every filter — or just the search, while you are typing in it

Ignored while you are typing. Pages without a filter rail have no search to focus.