Deviance of iter1=Inf

[AlanRace]

I am not quite sure why this isn't working for me, but I have essentially attempted to duplicate the iris example but with my own data and I get the following error...

[2022-07-26 13:28:01] Setting: epsilon = 0.2 and delta = 0.1


[2022-07-26 13:28:01] Initializing ROC-GLM

[2022-07-26 13:28:01] Host: Received scores of negative response

[2022-07-26 13:28:01] Receiving negative scores
  Aggregated (...) [=====================================================================] 100% / 3s
[2022-07-26 13:28:04] Host: Pushing pooled scores
  Assigned expr. (pooled_scores <- decodeBinary("580a000000030004010200030500000000055554462d3800...
[2022-07-26 13:28:07] Server: Calculating placement values and parts for ROC-GLM
  Assigned expr. (roc_data <- rocGLMFrame("D$binary", "pred", "pooled...
[2022-07-26 13:28:11] Server: Calculating probit regression to obtain ROC-GLM
  Aggregated (calculateDistrGLMParts(formula = y ~ x, data = "roc_data", w = "w", ) [====] 100% / 3s
[2022-07-26 13:28:14] Deviance of iter1=Inf
Error in if (stop_crit < stop_tol) { : 
  missing value where TRUE/FALSE needed

My guess is that update$likelihood is NaN, which in turn is setting lh to 0, and the subsequent log(lh) setting dev to Inf.

dsROCGLM/R/roc_glm.R

Lines 56 to 57 in 733eeb2

	lh = Reduce("+", lapply(update, function(x) ifelse(is.nan(x$likelihood), 0, x$likelihood)))
	dev = -2 * log(lh)

Any idea why update$likelihood should be NaN?

[schalkdaniel]

Thanks for pointing this out, we moved the package to a new repository dsBinVal. Please try the example there again and open an issue if it is still not working for you.