random forest (rf) for classification task

Note

for either regression or classification tasks using random forests or elastic-net, the y to be predicted is now named endpoint

Dependencies

check environment.yml

Data table format

data table should

1. have samples as row, and features as column
2. contain a column named endpoint (either class labels or continuous variable)
3. contain a column named sample_id - unique id for each sample/row

Run

1. navigate to the current directory of this folder /your/path/rf_classifier
2. run the main script with data table filename (e.g. demo_iris.csv) as input

Rscript main_rf.R demo_iris.csv

3. the trained model, intermediate files, and some output plots are stored under outs/[filename]

main_rf.R for classfication

1. configure: NUM_TREE (number of trees), RF_METRIC (metric to select the best model), PCT_PARTITION (percentage to split train, test data)
2. run Rscript main_rf.R [filename].csv

main_enet.R for classification

1. configure: FAMILY="binomial"
2. important for endpoint to have numerical entries, e.g. 1:"Survived", 0:"Died"
3. run Rscript main_enet.R [filename].csv

main_enet.R for regression

1. configure: FAMILY="gaussian"
2. endpoint is continuous, e.g. in demo_mpg.csv
3. run Rscript main_enet.R [filename].csv