Apply L0 regularization with Hard Concrete distributions to sparsify neural networks.
npx skills add https://github.com/PolicyEngine/policyengine-claude --skill l0
Please help me install this Agent Skill. Skill: l0 Source: https://github.com/PolicyEngine/policyengine-claude/tree/main/skills/l0-skill Command: npx skills add https://github.com/PolicyEngine/policyengine-claude --skill l0
This Skill helps PolicyEngine create more efficient survey datasets by intelligently selecting which households to include in calculations. This leads to faster population impact calculations and smaller dataset sizes while maintaining accuracy.
Use the l0 skill to select 1,000 representative households from your data DataFrame.
High-intent search queries and answers about installing and using this skill.
L0 regularization applies sparsity to neural networks by learning which weights and features are essential, automatically zeroing out less important ones. This reduces model size and computational cost while maintaining accuracy, using Hard Concrete distributions in PyTorch to approximate discrete selection decisions.
Yes. L0 regularization provides sampling gates to intelligently select representative households from survey datasets like CPS. This reduces the number of households needed for microsimulation while preserving statistical accuracy for population impact calculations.
L0 regularization identifies and removes redundant features and households, shrinking datasets and neural networks. Fewer computations mean faster microsimulation runs and quicker population impact analysis without sacrificing result reliability.
L0 regularization directly penalizes the count of non-zero weights, learning discrete feature selection through Hard Concrete approximations. Unlike L1/L2 which shrink weights gradually, L0 produces true sparsity by automatically eliminating entire features and gates.
Yes. L0 regularization integrates with PyTorch neural networks through l0-python dependency, enabling temperature-controlled sparsity and differentiable optimization across PolicyEngine's survey calibration and microsimulation pipelines.
L0 regularization requires careful temperature tuning to balance sparsity and model performance. Extremely large surveys may need experimentation to find optimal penalty weights and gate thresholds; overly aggressive sparsity can harm accuracy.
Imad-OuteBuild graph neural network models with PyTorch Geometric layers and datasets.
PolicyEngineAnalyze district-level policy impacts using PolicyEngine US microsimulation.
LynxLabVNTrain Sparse Autoencoders with SAELens to decompose neural network activations into interpretable features.
JunMA98Compute SHAP values and visualize feature attributions for model predictions.
