More gradients, more functions, more options

• Added IndGraph, the indicator of the graph of a linear operator;
• Added iterative version of IndAffine, LeastSquares, Quadratic, all accessible through the kwarg iterative=true in the constructor;
• Added gradient and subgradient computation for several functions;
• Several bugfixes and performance improvements.

See the diff.

Scaled packed storage in IndPSD

This minor update allows the packed storage representation of symmetric matrices in IndPSD to be used in scaled form, according to what described here.

See the diff.

Gradients and docs

• Methods gradient and gradient! now compute (guess what?) gradients in the case of smooth functions.
• The new type QuadraticIterative implements quadratic functions, just like Quadratic, but uses CG to perform the prox! operation.
• The new calculus rule Precomposition handles the general case where a function is composed with a linear operator L such that L*L' is diagonal.
• Added documentation.

New interface

Main changes:

• Updated argument order: from prox!(F, x, y, [gamma]) to prox!(y, F, x, [gamma])
• For separable functions, one can now use gamma::AbstractArray to use a different stepsize for each component
• Postcomposition is now Postcompose, Precomposition is now PrecomposeDiagonal

Functions added:

• Indicator of the rotated second-order cone: IndRotatedSOC

v0.2.1

more specific version cutoff

v0.1.0

Removed compatibility with Julia 0.4, added a few functions.

v0.0.1

fixed IndExpDual