ANNZ v2.2.0

• Added a bias correction procedure for MLMs, which may be switched off using glob.annz["doBiasCorMLM"] = False. (See README.md and scripts/annz_rndReg_advanced.py for details.)
• Added the option to generate error estimations (using the KNN method) for a general input dataset. An example script is provided as scripts/annz_rndReg_knnErr.py. (A detailed description is given in README.md.)
• Added the userWeights_metricPlots job option, which can be used to set weight expressions for the performance plots of regression. (See README.md for details.)
• Changed the binning scheme for the performance plots of auxiliary variables (defined using glob.annz["addOutputVars"]). Instead of equal-width bins, the plots now include bins which are defined as each having the same number of objects (equal-quantile binning). This e.g., reduces statistical fluctuations in computations of the bias, scatter and other parameters, as a function of the variables used for the training.
• Changed the default number of training cycles for ANNs from 5000 to a (more reasonable) randomized choice in the range [500,2000] (ANNZ::generateOptsMLM()). The option may be set to any other value by the user, using the NCycles setting. E.g., during training, set: glob.annz["userMLMopts"] = "ANNZ_MLM=ANN::HiddenLayers=N,N+3:NCycles=3500".
• Fixed minor bug in ANNZ::Train_binnedCls(), which caused a mismatch of job-options for some configuration of binned classification.
• Added a version-tag to all intermediate option files, with a format as e.g., [versionTag]=ANNZ_2.1.3.
• Minor change to the selection criteria for ANNZ_best in randomized regression.
• Other minor modifications and bug fixes.