List of possible low priority improvements

[Kertoo]

The following list describes possible upgrades to the package that would reduce memory usage/reduce computational complexity but are low priority since the functions to which they are connected already give correct results. They are suggestions for what to do when the main plans are done.

• Semiparametric bootstrap is slow because bootstrap sampling is done two fold, first the $y$ value is sampled from distribution $\mathbb{P}(y=k)=\frac{\hat{\boldsymbol{f}}_{k}}{\hat{N}}$ and then the vector of independent variables is drawn from units satisfying $y=k$. Doing these two steps simultaneously would decrease computational complexity. (Solved another way)
• Simulating Fisher information matrix for negative binomials could be faster, the argument $\ln\Gamma\left(y+\frac{1}{\alpha}\right) - \ln\Gamma\left(\frac{1}{\alpha}\right)$ is at fault a smarter way of computing $\mathbb{E}_{y}$ for these would significantly reduce both mu and cc.
• "Smart" way of iteratively changing momentum factor/step size in IRLS would speed up calculations possibly as an optional argument. Probably only matters for big datasets.
• Better handling of Xvlm and W matrixes in multiparameter IRLS. Eg by matrix package.

Some variables may be declared but redundant after updates (this should be routinely checked upon).

• Possible improvements to data storage in main function.
• Hatvalues are now computed directly which is computationally demanding. There is a way of obtaining them from QR decomposition which would be faster.
• "New" family functions have a terrible simulate element in them. Speed them up.