vismeteor

Breaking changes

• Both variance-stabilizing transformations were rebuilt. Arguments and usage are unchanged, but vmgeom_vst_from_magn() returns values on a different scale: 1.4 <= r <= 4 now maps to tm between about 1.52 and 3.44, where the old transformation gave 3.77 to 5.74. Transformed values and estimates stored with earlier versions are not comparable — recompute them from the raw magnitudes.
• vmgeom_vst_*() is documented for 1.4 <= r <= 4, previously 1.4 <= r <= 3.5, and stays usable outside it: at a limiting magnitude of 6.0 an r of 1.2 or 5.0 is now recovered to within 0.005, where the old transformation returned NA. Inside 1.7 <= r <= 3.3 the old one was somewhat more accurate — it carried a separate parameter set per limiting magnitude, which the single one used now replaces.
• dmideal(), pmideal() and qmideal() reject missing values in m/p and psi with stop("NA's are not allowed!"), consistent with qvmgeom(), qvmideal() and cvmideal(). Previously they either returned a missing value or failed with an unhelpful message. Infinite magnitudes remain valid.

Changes

• psi = Inf is now a valid value throughout. It denotes the geometric limit the ideal distribution converges to and is accepted by dvmideal(), pvmideal(), qvmideal() and rvmideal(), which previously rejected it with stop("Infinite psi values are not allowed!"). vmideal_vst_to_psi() and predict() return it where the data cannot distinguish a larger psi from a smaller one — a case that used to yield NA — so such an estimate can be passed on unchanged instead of being intercepted.
• vmgeom_vst_to_r() no longer returns NA outside a calibrated window. It now rejects only values the model cannot produce at all, so sparse data in a predictive model yield an implausibly large but correctly ordered r rather than a dropout.
• dvmideal(), pvmideal() and qvmideal() now honour a perception_fun given to them also for psi = Inf, where they previously fell back to vmperception().
• The warning issued by qmideal(), qvmgeom(), qvmideal() and cvmideal() reads NAs produced instead of NaNs produced, matching what these functions actually return. Code that filters warnings on the exact message needs updating.