Function Sets

[daniel1noble]

The sets of functions
Preparation functions

• Check class of columns
• Prior functions - sets prior based on class, variable and random effect
• Function assessing amount of missing data in each column
• Subset predictors/ columns function. In case user specifies the specific columns they want to impute.
• Formula building functions. Random effects with and without species. And fixed effects.

Statistical functions

• Predict functions for various error distributions
• Model fitting function that builds model
• Trace/convergence function from T0 to Tn for response variable
• Convergence of MCMC runs, autocorrelation that depends on continuous vs categorical variable

Reporting functions

• As its running we want to report the cycle and information on x1, x2, x3, y as it progresses
• Methods matrix summary function
• Two types of reporting. One that just reports on the last iteration, but provides more detail. Others that report on each cycle as the function progresses.
• Cross validation. Manually reduce known data and impute known data to see if it recovers the known data.

[itchyshin]

@daniel1noble and @szymekdr

There are a number of adoptable/adaptable functions in hmi. For example, they have imputation functions for: 1) Gaussian, 2) Count, 3) Binary, 4) Ordered Nominal and 5) Unordered Nominal (prior etc can be borrowed from these). Also, @daniel1noble we should look at their formula building together.

To modify these I need to:

• 4 types of data, Y, X, Z (only one slope allowed) and Cluster (species)
• Each function can have the default iterations etc (e.g. nominal one will take for ever long time).
• Using these functions - we are imputing allowing all sorts of errors so we will need to cycle a lot but final model with M > 10 should capture all uncertainty (not sure what to do with predictors yet)
• They do have a function to sort out which type of data (e.g. continuous, binary etc) so I can rewrite this

That is all for now - and I will keep adding as they come up