An R package containing functions I've written while working on predictive modeling projects. The functions are briefly described below, but documented in more detail in the function-specific help (.rd) files. All content can be downloaded/installed and then used like any other package after following the installation instructions below.
Install via devtools straight from GitHub:
install.packages("devtools")
library("devtools")
install_github("etlundquist/eRic")
Download the tarball eRic_0.0.0.9000.tar.gz and install from source:
install.packages('/path/to/tarball/eRic_0.0.0.9000.tar.gz', repos = NULL)
- bootImp - calculate variable importance/perform feature selection using bootstrap resampling and a combination of different filter/model-based methods. Available methods are:
- Information Value
- Chi2
- Random Forest
- GBM
- Bagged Earth/MARS
- Bagged Lasso
- evThreshold - calculate an optimal probability threshold for classification given the costs/benefits for each confusion matrix cell
- provides an estimate of the optimal probability cutoff with respect to confusion matrix utility
- provides an estimate of unit expected value given your model and the optimal probability cutoff
- varCluster - cluster predictor variables and extract cluster summaries for dimension reduction
- use agglomerative clustering (hclust) to group highly correlated sets of predictor variables
- create a single variable to summarize each cluster (highest pairwise, highest x-y, centroid, PC1)
- produce a correlation plot to visualize correlation structure in predictor matrix
- plotYXbin - produces a ggplot object with bin values of Y with respect to bins of X
- Y metrics include: [proportions, log-odds, WOE]
- X split methods include: [quantile-based, uniform splits, rpart-based splits]
- Can specify the desired number of bins and whether a missing value bin should be added
- Additionally calculates Information Value (IV) and Chi2 Statistic for the XY relationship
- prCalibrate - perform Platt Scaling on raw model predicted probabilities to better align with actual class proportions and produce a calibration plot to visualize results
- scale predicted probabilities with respect to either a calibration or independent validation set
- produce a calibration plot showing the relationship between actual and predicted class proportions
-
plotKS - produces a ggplot object with a KS plot for two distributions as well as the KS statistic value
-
tsSummary - calculate summary statistics (using passed summary functions) with respect to wide-format time series variables
-
tsTrends - calculate linear trends with respect to wide-format time series variables