Build me R code to test all possible regression specifications

I work in research economics and am looking to automate some of my modeling. I know this requires a set of for loops and some programming/stats knowledge but I don't have the programming skills. I want to use this mainly for inference/explanation but most machine learning tools I could find focus entirely on prediction. I am looking for a program in R that does the following: Step 1) I put in my data and select a dependent variable. Step 2) It splits data into multiple randomly chosen training and prediction sets. Step 3) It runs a regression with each possible combination of independent variables (for example, if I have income I as a dependent variable and education ED, experience EX, and gender G, it runs a regression for each of ED, EX, G; ED, EX; ED, G; EX, G; ED; EX; G; against I) -- a) It would be a bonus if it can also do this but with each combination of their logs (I against log EX, ED, and G; I against log EX, log ED, and G, log I against EX, log ED, and so on) and their exponents up to 3 (I against EX^2, EX, ED, or I against EX^2, ED^2, EX, ED, G, and so on). Step 4) As an output, it gives me the total/overall/combined (for multiple training and prediction sets) stats for each model specification in this format: a table (or list, but table format would be easier to read) with each model specification used and its R-squared, adjusted R-squared, Akaike info criterion, Schwarz info criterion, and ideally also the p-values for all its variables. I will mostly be using this for cross-sectional or panel data, not time series, so the training and prediction sets can be any selection. Just regular OLS regression is alright, but if it tests a bunch of other options as well or even lets me select which kind of regression to run that would be a bonus.