Visualizing Machine Learning of Units of Measure using PyViz
summary
Claims that ClinicalTrials.Gov yields 35,926 trials with results but 24,548 distinct unit strings, so the project builds a pipeline that clusters and maps those variants and serves them on ClinicalUnitMaping.Com. It uses PyViz visualizations (HoloViews + Datashader interactions) to inspect k‑means clusters and to debug sequence models, and trains supervised architectures (seq2seq LSTM/CNN variants in Keras/TensorFlow). Prototype mapping already yields ~56% top prediction accuracy and ~78% within 10 tries.
AI-written summary · about
chapters (12)
- 0:00Visualizing ML units
- 1:03Non-standardized units in medical data
- 3:31Computers and medical numbers
- 5:20Computer problem-solving milestones
- 8:10Extracting unit standards for healthcare
- 10:20Unsupervised learning for unit clustering
- 14:16Building unit mapping website
- 18:16Supervised ML for unit correction
- 21:08Testing neural networks for unit prediction
- 24:17Debugging neural networks with PyViz
- 26:06Summary and future of unit standardization
- 28:57Q&A on unit standardization
related talks
43mVisualising Machine Learning Models
03h 51mMachine Learning Summit
26mExtracting Medical Attributes from Clinical Trials
39mVisual diagnostics for more informed machine learning
30mProductionizing an unsupervised machine learning model to...
35mVisual Diagnostics for More Effective Machine Learning
50mBuilding Intelligent Mobile Apps using Machine Learning
19mHow to look at your data