Adjustments to Aphrodite functions

[kniehaus]

Major changes:

• Some of the feature processing (lab features) wasn't quite functional. Now seems to be working.
• The model building portion wasn't quite functional and also seemed to be creating a regression model, rather than classification. This portion of the code runs for me in RStudio but not from the terminal, so perhaps there are still problems.
• The feature manipulation has been sped up quite drastically by using data table (vs. data frames). This is most noticeable with large sample sizes.
• Wrote a function to look at the feature weightings. The top features suggest to me that there still may be bugs in the code...

NOTE: still learning R so if something seems to not make sense, it probably doesn't - let me know!

[kniehaus]

Major changes:

• Added in filtering of features to only include those that: do not fall into non-useful domains (from what I can tell), are standard terms, and not invalid, and are not included in the list of keywords used to define the patient cases class
• Added in option to look at features only before first keyword appearance, or only after
• Normalized feature counts by the the length of patient follow-up (e.g. if 6 observations of term Y, seen over a 2 year period, feature Y normalized to 3 for this patient)
• Altered usage of ignorelist_ed: instead of a single ignore list of keywords, created ignorelist_caseIDs (which include keywords that we want to use to exclude patients from the cases class) and ignorelist_features (which include features that we want to exclude as features)
• Split part of buildModel function that combined the feature sets into a separate function: combineFeatureVectors. Also added in option for filtering out features that appear in less than x% of the patients
• Split sample code into a "pre-keywords" portion and a "post-keywords" portion
• Added in random forest option for classifier (still just using caret)

A couple to-dos:

• introduced some bugs into the lab terms portion - need to address these
• add in option to center/scale the data (not quite clear yet on how caret does this - whether inside cv training folds, or just applied as blanket to entire dataset pre-model building) (others as noted throughout the code)

[kniehaus]

Hopefully all options that I changed are now available as flags, and lab feature portion should be fully functional. I haven't included an updated example script here, but can after I clean it up a bit.