EuroForMix v3.0.0 (Release date: 2020-04-01)
=============================================
Summary of version 3 updates:
- Faster calculations due to paralellization on deeper level; causes speed to scale with the number of logical CPU cores (threads).
- Progress bar for all types of calculations. Time estimates given for the MLE method (if time exceeds 10s).
- Automatic model selection for the MLE method (number of contributors, Degradation (ON/OFF), BW-stutter (ON/OFF), FW-stutter (ON/OFF) decided based on the AkaikeInformationCriterion)
- Model extensions: Forward stutter model and Marker/Dye specific settings for analytical thresholds, dropin model and fst.
- Validation plots created (almost) instantly
- Seed settings: The MCMC and conservative LR values are now reproducible. Also optional seed can be given for the MLE method (reproducible start values, useful for time comparisons).
- Upper boundary of Likelihood Ratio provided in GUI (1/'Random match probability').


Details of the update:
 - GUI changes (efm):
	- In 'Generate data' panel: New layout. Includes Forward stutter. 
	- In 'Model specification' panel: Possibility of forward stutter model added.
	- In 'MLE fit' panel: New layout for better per-marker space. 
		- 'Only LR results' removed. 
		- adj.loglik added to loglik: adj.loglik = loglik - nparam
	- In create report this was added: 1) Frequency for rare alleles and 2) information of whether the frequencies were normalized after imputing rare alleles.
	
 - Bug fixes:
	- In function deconvolve: The genotype of known contributors were not taken into account when doing deconvolution with theta-correction (only known non-contributors).
	- In function plotTopMPS2: Inserting base pair of closest allele (for missing bp)

 - New implementation of the C++ engine (Armadillo/Rcpp not used any more):
	 - The C++ engine performs parallelization on genotype marginalisation over one for-loop which makes the speed scale with the number of CPU cores.

 - Model extensions:
	 - Forward stutter model.
	 - Marker dependent AT/pC/lambda/Fst parameters (also implemented for qualitative model).

 - New features:
 	- Monitoring progress (a progressbar is provided in console)
	- Automatic model selection (based on the AIC): Number of contributors, Degradation (ON/OFF), BW-stutter (ON/OFF), FW-stutter (ON/OFF)
 	- Seeds added to following functions: contLikMCMC,contLikMLE 
	- Optional 'normalization' handling of rare alleles (missing alleles in population frequency data): 
		- Can be changed under "Frequencies -> Set whether to normalize frequencies" (1=Yes, 0=No) 
		- Allele frequencies are by default normalized after including new alleles. In EFM v2 the default was not to normalize - set option equal 0=No for concordant results.

 - Following functions has been modified and updated because of the new C++ implementation:
	- prepareC
	- contLikMLE
	- contLikINT
	- contLikMCMC (depends on contLikMLE output)
	- validMLEmodel (depends on contLikMLE output)
	- deconvolve (depends on contLikMLE output)
	- logLiki (depends on contLikMLE output)

 - Following functions are not used anymore (but still there): contLikMCMCpara (calls contLikMCMC), contLikMLEpara (calls contLikMLE), iszerolik

 - Added functions: 
	- runexample: Illustration of using the most important methods in euroformix.
	- calcRMP: Calculates the random match probability of a specific profile.
	- conLikSearch: Searches through all model outcomes and shows the optimal model in the MLE panel. Table is visualised and also stored in report file.
	- sample_listToTable: Converts list structure to table (used by exporting)
	- sample_tableToList: Converts tables to list (used by importing)
	- prepareData: preparing data with Qassignation and filter data wrt analytical thresholds (possibly per marker)

 - Modified functions:
	- genDataset: Also includes forward stutter simulation and marker based analytical threshold
	- getKit: The user can specify other kit files than the one by default (more flexibility)
	- simDOdistr: The function now accepts dropin probability per marker (if provided)
 
 - Includes automated testing of the numerical calculations by C++ using 'testthat' (compares with manual R implementations)
	- See separete presentation "EFM3numtestPres" for a description
 
 - Other notes:
	- Degradation slope param can never exceed 1. This was possible before 
	- New optimization strategy for MLE (contLikMLE): 
		-> New strategy: Require nDone number of obtained local maximum values (set to 2 as default).
		-> Random start points are better chosen: Mixture proportions (flat, but sorted for unknowns), PHexp/PHvar/stuttprop/degradslope have lower variation.
		-> Faster convergence in nlm, but about same accuracy (minimum step tolerance (steptol) set to 1e-3 instead of 1e-6).
	- New sampling strategy for MCMC: No block-wise sampling provided and only one chain is run.
	- The parallel versions contLikMLEpara and contLikMCMCpara are no longer used (calls the original functions)
	- The likelihood of underflow numbers are now shown in MLE results.