big-life-lab/bllflow

Big Life Lab Flow (BLLFlow) - a workflow for open, reproducible research. Includes support for PMML, DDI.

bllflow's Issues

Packrat

Examine Packrat and whether it can help with dependencies issues with moving the package across firewalls. @yulric @StaceyFisher putting this on your radar as well. https://rstudio.github.io/packrat/

Change main data examples to pbc in survival package (primary biliary cirrhosis)

Fix warning for non-standard folder

Image folder is non-standard.

Possible fix: make a vignette to explain add-in use. (within Vignette folder).

What to do with summary table attached to bllFlow

How should the table attached to bllFlow be modified (appended or merged)

brier_score followup time

Change brier_score function so that followup doesn't necessarily need to be called 'FollowUp'

v0.1.2 checklist

• Rename repo to bllflow
• Update version in DESCRIPTION to 0.1.2

make_event_dates units

The make_event_dates function isn't working for me in my branch or in master.

The error I am getting is:
Error: 'units<-' is not an exported object from 'namespace:Hmisc'

After looking at this post I tried updating my Hmisc and dplyr packages.... I'm still getting the same error:
sparklyr/sparklyr#1703

Anyone else?

Summary table

Allowing option to merge multiple summary tables as well as simply appending to tables.

DDI survey data for bllFlow

Here is a starting list of metadata that can form the basis of the DDI object within bllFlow. I've put a copy of the DDI document in the PoRT repo. For bllFlow examples, we are using the pbc data. Currently, the pbc data doesn't have a DDI file, so we'll need to use the CCHS file as the example, until we create a DDI file for the pbc data.

https://github.com/Big-Life-Lab/PoRT/blob/master/CCHS_DDI/CCHS-82M0013-E-2013-2014-Annual.xml

Unnecessarily complicated example for brier_score

It is difficult to write a example for the brier_score function with the currently available sample datasets/ dataset creation functions. Brier score for survival requires calculation of the probability of censoring across time to create weights- a dataframe created using RESPECT-EOL_validation.csv and 'make_event_date' function results in a df that has no censoring until the last time point (i.e 0% censor for all time points up to time of interest, then 100% censor). This results in infinity as the brier score, if the time of interest is the same as the time of censoring. Can't easily use a different value as the time of interest, as the available values are generated randomly with every run of 'make_event_date'.

I've figured out a round-about way to solve this, by having the time of interest be the time nearest to, but less than censoring time. However, this makes for an unnecessarily complicated example.

Best solution would be to use a more realistic dataset that includes censoring throughout time, rather than just at end time.

Labels across R package

Let's review and discuss our approach to:

• Importing labels and metadata from the Model Specification Worksheet.
• generating labels when we clean observations and/or transform variables.

Search not working