Abnormal termination in some optimization cases

[jcarlis3]

@tmcd82070 Using example sparrow data and the example below, there are issues in the optimization for one covariate (bare), but not another (shrub, which happens to have strong inverse correlation to bare).

Produced by code below: Warning message: In F.dfunc.estim(formula = dist ~ bare, data = sparrow.merge, likelihood = "halfnorm", : ERROR: ABNORMAL_TERMINATION_IN_LNSRCH

---

Prep reproducible example with package data. data(sparrow.detections) data(sparrow.sites) sparrow.merge <- merge(sparrow.detections, sparrow.sites, by="siteID")

No problems when shrub is the covariate good <- F.dfunc.estim(formula=dist~shrub, data=sparrow.merge, likelihood="halfnorm", w.hi=150)

Convergence failure when bare is the covariate bad <- F.dfunc.estim(formula=dist~bare, data=sparrow.merge, likelihood="halfnorm", w.hi=150)

---

Some leads:

• https://stackoverflow.com/questions/34663539/scipy-optimize-fmin-l-bfgs-b-returns-abnormal-termination-in-lnsrch
• http://r.789695.n4.nabble.com/optim-method-quot-L-BFGS-B-quot-abnormal-termination-td819101.html

[jcarlis3]

Now, a different error is returned:

The reproducible code above was before renaming some functions -- see below for current.

data(sparrowDetectionData) data(sparrowSiteData)

good <- dfuncEstim(formula=dist~shrub, detectionData=sparrowDetectionData, siteData=sparrowSiteData, likelihood="halfnorm", w.hi=100)

bad <- dfuncEstim(formula=dist~bare, detectionData=sparrowDetectionData, siteData=sparrowSiteData, likelihood="halfnorm", w.hi=100)

[jcarlis3]

For the time being, we recommend standardizing problematic covariates, which appears to help. We're working to stabilize optimization in another issue.