In cvxgrp/CVXR: Disciplined Convex Optimization

What is new?

Version 1.0 is a major upgrade. The improvements include:

• A new reduction framework that automatically chooses the most appropriate solver for a problem
• Addition of a large number of solvers using native R interfaces
• Facilities for geometric programming
• Improvements in speed.

What has changed?

Implementation changes include:

• Default QP solver: now, OSQP, which means that some results will differ slightly from previous runs. However, ECOS can be explicitly specified if exact replication of old results are desired
• Strict inequalities are not allowed in constraints.

Syntax changes (to match cvxpy 1.x) include:

• Int(m, n) is now Variable(m, n, integer = TRUE)
• Bool(m, n) is now Variable(m, n, boolean = TRUE)
• Semidef(n, n) is now Variable(n, n, PSD = TRUE)

and so on. Details may be found on the CVXR Website.

cvxgrp/CVXR documentation built on Nov. 8, 2024, 5:47 p.m.