Tuning Hyperparameters with DVC Experiments
summary
Grid search with three hyperparameters and ten values each explodes to 1,000 runs while random search often finds better settings in the same time by sampling broadly. DVC actually bundles code, data, params and metrics into reproducible experiment snapshots stored in the DVC cache and referenced via Git metadata. Commands like dvc exp run and dvc exp show queue, run (serial or parallel) and list hundreds of experiments, and plots diff produces quick metric comparisons so specific experiments (for example #71) can be reproduced exactly.
AI-written summary · about
chapters (8)
- 0:00Introduction to ML experiment tracking
- 3:48Common issues in hyperparameter tuning
- 8:49DVC for experiment management
- 15:25Setting up experiment queues
- 18:33Reviewing experiment results
- 21:03Visualizing experiment metrics
- 26:42Benefits of reproducibility
- 28:29Key takeaways and wrap-up
related talks
38mUsing Reproducible Experiments To Create Better Models
16mMake your data science workflows reproducible
46mFlexible ML Experiment Tracking System for Python Coders with DVC and Streamlit
33mFlexible ML Experiment Tracking System for Python Coders with DVC and St...
27mLeveraging Git for ML Experiment Management
37mOpen Source Tools for ML Experiments Management
02h 10mBuild End-To-End MLOps Platform with Open Source DVC Ecosystem
25mHyperparameter Tuning and Visualization in Deep Learning