convexopt

Convex optimization framework for Python

TODO

• tests, examples, documentation
  • scipy's nosetests with np autoimport
  • build documentation using Sphinx
  • make sure complex numbers are handled correctly
• constraints (proximal mapping is orthogonal projection)
  • PositiveIndicatorFunction
    • gradient: PositiveIndicatorFunctionGradient
      • backward: return np.maximum(x, 0)
  • how can constraints be combined with other functions?
    • e.g. ||x||_1 s.t. x >= 0
• stacked and reshaped operators
• inspection to find out what methods are implemented
• automatically apply the Moreau decomposition: x = prox_f(x) + prox_f*(x) => x = f.gradient.backward(x) + f.conjugate.gradient.backward(x) => If f.gradient or f.gradient.backward are not implemented, try f.conjugate.gradient.backward instead.
• logging to console using the logging module
• cached sparse decomposition for DataTerm backward operator