Casimir API summary

Optimization

Incremental First Order Oracles - Base classes:

casimir.optim.IncrementalFirstOrderOracle()	Base class for incremental first order oracles (IFO).
casimir.optim.SmoothedIncrementalFirstOrderOracle([...])	Base class of smoothed incremental first order oracles.

Optimization Algorithms:

casimir.optim.optimize_ifo(initial_model, ...)	Minimize a convex function with access to a (smoothed) incremental first order oracle using a primal algorithm.
casimir.optim.CasimirSVRGOptimizer(...[, ...])	Implement Casimir (Catalyst with smoothing) or Catalyst outer loop with SVRG as the inner loop.
casimir.optim.SGDOptimizer(initial_model[, ...])	Implement the stochastic (sub-)gradient method with various learning rate and averaging schemes.
casimir.optim.SVRGOptimizer(initial_model, ...)	Implement Stochastic Variance Reduced Gradient (SVRG) with optional smoothing.
casimir.optim.block_coordinate_frank_wolfe_optimize(...)	Implement the Block Coordinate Frank-Wolfe (BCFW) algorithm for structured prediction.

Regularization:

casimir.optim.L2Penalty(regularization_parameter)

Class representing the L2 regularization penalty.

Data

Classification:

Named Entity Recognition:

Object Localization:

Structured Prediction Utilities