bcjaeger / bimm Goto Github PK

make sure outcome is binary, if not give error message.
no missing data, if missing any values give error message.

bimm_validate(predictions from bimm, observed values, prob cutpoint)
create table with AUC, brier, sens, spec, NPV, PPV and Confidence intervals
not necessary, let's think about this more

bimm_predict() with a test dataset of new_obs (new observations) outputs odds ratios, but it should output predicted probabilities to be consistent with the bimm_predict() for test data of new subjects. can you please edit this?

I am attempting to read through your code and possibly use BiMM for some of my own research. However, I have been unable to run the examples shown in your ReadMe file. I received the following errors:

Beginning iteration 1...mixed effects model could not be fit; ending on iteration 1
Warning messages:
1: In get(optimizer) : restarting interrupted promise evaluation
2: In get(optimizer) : internal error -3 in R_decompress1

I also attempted to run this on an example dataset of my own and I received the same warning/error messages from bimm_fit.

Is there a way to make it run for an unspecified number of iterations until the log likelihood converges? Instead of having the user specify this?

we need a function that takes a test dataset of new observations for subjects included in training data that calculates the predicted probability from the ML model ._x_. and uses this in the model_mer to predict the outcome.

using summary to write more expressive prints:
output from RF
output from GLMM

make two predict functions:
predict_new(), predict_newsubj(),
versus
predict_within(), predict_newobs()

statmod should be added to dependencies

I was trying to run code for the multiple iteration version of BiMM forest where it goes until it converges. However, I am getting an error message that I need to specify the n_iteration in the bimm_fit function. can you please look into this?