from an e-mail exchange with @drizzersilverberg:

1. Can 'sgd' package perform linear regression task to build a model and predict a set of data using the built model?

2. If so, can you show me the way (code/script) to build the model and predict simple data? as an example, you can use this house_price data.

3. Last, can 'sgd' compute the prediction error using mean-square-error/root-mean-square-error? If > so, can you show me the way (code/script) to use it in above example?

my reply:

1. the main function sgd will estimate parameters for a chosen model such as linear regression. there are utility functions to handle the tasks post-estimation. for example, the predict function takes the output of sgd and test covariates/features as input; the output is the predicted response. (see ?predict.sgd)

2. following the example on the README.md (https://github.com/airoldilab/sgd/blob/master/NAMESPACE), you can do something like:

X_test <- matrix(rnorm(50*d), ncol=d)
y_hat <- predict(sgd.theta, cbind(X_test, 1))

3. to get the numerical values, you have to do it manually; for example, we have a demo for mean-squared error over the parameters (see ?sgd):

sprintf("Mean squared error: %0.3f", mean((theta - as.numeric(sgd.theta$coefficients))^2))

however, we also have plots that can do MSE or classification error in predictions. (see ?plot) including the numerical output would be as simple as exposing the utility functions we wrote there; we chose not to in order to force the user to not rely on helper functions.

i think these could be made more explicit in the documentation.