Release notes for sbmlsim 0.2.0

• breaking changes in units handling
• bugfix plotting dimensionless units in labels
• remove deprecated distrib and sampling functionality.
• fixed typo in sensitivity

parameter fitting

• refactoring parameter fitting
• improved weight and options handling
• documentation
• reports for parameter fitting (fit mapping contribution, #79)
• storing and combination of parameter fitting results (#73)
• better cost plots (#72)
• improved parameter fitting results plots (#96)
• loss functions implemented (#99)
  • ‘linear’ (default) : rho(z) = z. Gives a standard least-squares problem.
  • ‘soft_l1’ : rho(z) = 2 * ((1 + z)**0.5 - 1). The smooth approximation of l1 (absolute value) loss. Usually a good choice for robust least squares.
  • ‘cauchy’ : rho(z) = ln(1 + z). Severely weakens outliers influence, but may cause difficulties in optimization process.
  • ‘arctan’ : rho(z) = arctan(z). Limits a maximum loss on a single residual, has properties similar to ‘cauchy’.
• figures closing in reports (#88)
• improved parameter fitting plots and results (#32)
• bugfix all NaN errors (#101)