End to End Data Science Without Leaving The GPU
summary
Claims an end-to-end GPU workflow using JupyterLab, Ibis, the OmniSci Jupyter kernel and Apache Arrow zero-copy Arrow references so data never serializes off the device; demonstrates forecasting electricity demand with a simple linear model solved via GPU linear algebra in 497 microseconds and an interactive map rendering 42 million points live on the GPU; notes a visualization step that transfers a small sample back to CPU as a practical cheat and explains where the pipeline currently has rough edges.
AI-written summary · about
chapters (7)
- 0:00OmniSci and GPUs intro
- 3:54GPU Open Analytics Initiative
- 5:00GPU data exploration demo
- 12:29End-to-end GPU workflow
- 16:22GPU forecasting model
- 24:56Future of GPU analytics
- 28:10Q&A on GPU memory
related talks
43mAdvanced Data Visualizations In Jupyter Notebooks
26mWho Said Wrangling Geospatial Data at Scale was Easy?
32mAdvanced Visualization in Jupyter Notebooks
35mClimate change analyzing remote sensing data with Python
33mfastplotlib-scientific discovery through data visualization
27mBuilding web-based engineering applications with JupyterLab components
32mLightning Talks
46mA 101 in Time Series Analytics with Apache Arrow, Pandas and Parquet