NEWS.md

In andrija-djurovic/PDtoolkit: Collection of Tools for PD Rating Model Development and Validation

PDtoolkit 0.1.0

Changes:

1. The new argument added to stepMIV function - coding.start.model which allows user to have different coding types for starting and final model. Additionally, the same function is improved adding the check for its output value - if (nrow(steps) > 0) {steps <- cbind.data.frame(target = target, steps)} and correction for miv table for missing/infinite values is introduced.
2. Improvement of cat.bin function is performed. If merging of special case bins is selected (argument sc.merge), then summary table output reports the bin with which it is merged.
3. Package is extended with a new functions: psi and create.partitions.

PDtoolkit 0.2.0

Changes:

1. rf.clustering - increased number of maximum clusters from 30 to 100 for manual selection. For x2y metric, minsplit and minbucket added in order to speed up the algorithm.
2. segment.vld - correction for possible 0 and 1 observed default rate in the prop.test.
3. replace.woe - extended list of elements for WoE check (c(NA, NaN, Inf, -Inf)).
4. The new argument added to stepMIV function - offset.vals. The same function, now returns the model development database also for coding = "dummy".
5. Package is extended with a new functions: evrs and interaction.transformer.

PDtoolkit 0.3.0

Changes:

1. Package is extended with a new functions: stepFWD and stepRPC.

PDtoolkit 0.4.0

Changes:

1. Package is extended with a new functions: staged.blocks, embeded.blocks and ensemble.blocks.
2. Fixed bug in create.partitions function - risk factors with more than 10 modalities.

PDtoolkit 1.0.0

Changes:

1. psi** value added to the output of psi function (for comparison with cv.zscore`` andcv.chisq``` critical value)
2. Improvement of cat.bin output consistency for sc.merge option
3. Additional check of segmentargument in homogeneity function (has to be of length one)
4. Function segment.vld parameterized with the new argument min.leaf
5. Additional condition considered for selection of next entry in stepFWD (now AIC value can be possibly considered in the selection process)
6. Bug fix in interaction.transformer function - identification of upper bound for partitioning
7. Argument sc in the functions of univariate analysis extended for -Inf value
8. New functions:
   • helpers functions: num.slice, cat.slice and encode.woe
   • nzv - near-zero variance
   • smote - Synthetic Minority Oversampling Technique
   • constrained.logit - constrained logistic regression
   • rf.interaction.transformer - extract interactions from random forest
   • hhi - Herfindahl-Hirschman Index
   • normal.test - Multi-period predictive power test
   • confusion.matrix and cutoff.palette - confusion matrix analysis
   • ush.test and ush.bin - U-shape testing and binning procedures
   • kfold.idx - indices for K-fold validation
   • fairness.vld - model fairness validation
   • decision.tree - custom decision tree algorithm and its predict method

PDtoolkit 1.0.1

Changes:

1. print from within the functions (stepMIV, stepFWD, stepRPC, staged.blocks, embedded.blocks, ensemble.blocks) replaced with messsage
2. Examples modified (stepMIV, boots.vld, segment.vld, scaled.score, kfold.vld, fairness.vld, evrs, staged.blocks) to keep the execution time under 10s during check_win_release()

PDtoolkit 1.1.0

Changes:

1. imp.outliers function did not replace identified outliers properly. Small adjustment made (db[, rf.l] <- rf.imp have been added).
2. nzv - label of the second most frequent values was wrongly assigned. (cc.lbl.2 = x.cc.lb1 replaced by cc.lbl.2 = x.cc.lb2)
3. rf.clustering - updated link for x2y metric
4. improvement of trend check (cc.dummy) in stepwise regressions
5. New functions:
   • stepFWDr - stepwise regression for mixed risk factor types
   • stepRPCr - stepwise regression based on risk profile concept and mixed risk factor types
6. New method available for staged.blocks, embedded.blocks, ensemble.blocks - "stepFWDr" & "stepRPCr"
7. Examples modified (staged.blocks, embedded.blocks, ensemble.blocks, rf.clustering, hhi, evrs) to decrease the execution time during check_win_release()
8. stepFWD and stepRPC - additional check for dummy coding and check.start.model introduced
9. rs.calibration output exteneded. Now, besides calibrated values it returns also parameters

PDtoolkit 1.1.1

Changes:

1. cat.bin function adjusted in part after dealing with special cases.
2. psi - typo in helper function num.bt corrected (instead of incluse.lowest = TRUE, now include.lower = TRUE), This change should not affect previous usage of the psi function because argument breaks is a single number which already ensures inclusion of extreme values (for details see ?cut)
3. The following helper functions renamed to avoid S3 method mismatches: tbl.correction (tbl_correction), summary.tbl (summary_tbl), log.likelihood(log_likelihood), best.split.num(best_split_num), best.split(best_split), best.split.cat(best_split_cat), sum.adjacent (sum_adjacent), c.best.split.num(c_best_split_num), c.best.split(c_best_split), c.best.split.cat(c_best_split_cat)

PDtoolkit 1.2.0

Changes: 1. pp.testing - description of the Hosmer-Lemeshow test results changed 2. power - for the Hosmer-Lemeshow test removed condition which checks if the observed portfolio default rate is less than predicted one.

PDtoolkit 1.2.1

Changes: 1. pp.testing - typo in the part of the arguments checks (is.numeric(n) converted to is.numeric(no)). This change does not impact the function output. Erorr message adjusted (instead of stop("All arguemnts have to of numeric type.") new message is stop("Arguments pdc, no, and nb have to of numeric type.")) 2. fix in rf.interaction.transformer function when tree cannot be generated based on the selected criteria

andrija-djurovic/PDtoolkit documentation built on Jan. 29, 2024, 9:35 a.m.