Optimization Advice
This section is meant to provide some guidance if you have trouble with root-finding or minimization for a given set of estimating functions.
A weakness of delicatessen is that it does not have the fastest or most robust routines for estimating statistical
parameters. This is the cost of the flexibility of the generic estimating equation implementation used (a cost that
delicatessen presumes to be worthwhile when being applied).
Below are a few recommendations for solving \(\theta\) with MEstimator or GMMEstimator that have been used
to some success in computationally demanding problems previously.
Center initial values
Optimization will be improved when the starting values (those provided in init) are ‘close’ to the \(\theta\)
values. However, we don’t know what those values may be. Since the optimization procedure needs to look for the best
values, it is best to start it in the ‘middle’ of the possible space. If initial values are placed outside of the
bounds of a particular \(\theta\), this can break the optimization procedure. Returning to the proportion,
providing a starting value of -10 may cause trouble, since proportions are actually bound to \([0,1]\).
For many regression problems, starting values of 0 are likely to be preferred in absence of additional information
about a problem.
To summarize: make sure your initial values are (1) reasonable, and (2) within the bounds of the possible values of \(\theta\), and (3) close to the actual estimate (if known or can be approximately known).
Pre-wash initials
In the case of stacked estimating equations composed of multiple estimating functions (e.g., g-computation, IPW, AIPW),
some parameters can be estimated independent of the others. Then the pre-optimized values can be passed as the initial
values for the overall estimator. This ‘pre-washing’ of values allows the delicatessen optimization to focus on
values of \(\theta\) that can’t be optimized outside.
This pre-washing approach is particularly useful for regression models, since more stable optimization strategies exist
for most regression implementations. Pre-washing the initial values allows delicatessen to ‘borrow’ the strength of
more stable methods. A pre-washing procedure that does not require solving for all coefficients is to instead set the
intercept at the mean of the outcome variables (or its transformed variation when using a generalized linear model).
Finally, delicatessen offers the option to run the optimization procedure for a subset of the estimating functions
via the optional subset argument. Therefore, some parameters can be solved outside of the procedure and only the
remaining subset can be searched for. This option is particularly valuable when an estimator consists of hundreds of
parameters.
Increase iterations
If neither of those works, increasing the number of iterations is a good next place to start. The default is 5000
but can easily be increased via the maxiter optional argument in the estimate() function.
Different optimization
By default, MEstimator uses the Levenberg-Marquardt for root-finding and GMMEstimator uses the BFGS
for minimization. delicatessen also supports other algorithms available in scipy.optimize. Additionally,
custom or manual root-finding algorithms can be used. Some algorithms may have better operating
characteristics for particular types of problems.
Non-smooth equations
As mentioned elsewhere, non-smooth estimating equations (e.g., percentiles, positive mean deviation, etc.) can be
difficult to optimize. In general, it is best to avoid using delicatessen with non-smooth estimating equations.
If one must use delicatessen with non-smooth estimating equations, some tricks we have found helpful are to:
increasing the tolerance (to the same order as \(n\)) or modify the optimization algorithm.
A warning
Before ending this section, I want to emphasize that simply increasing tolerance is not generally advised. While
it may allow the optimization routine to succeed, it only allows the error of the optimization to be greater.
Therefore, the optimization will stop ‘further’ away from the zero of the estimating equation. Do not use this
approach to getting delicatessen to succeed in the optimization unless you are absolutely sure that the new
tolerance is within acceptable computational error tolerance for your problem. The default tolerance is 1e-9.