[Andrea Smurra]

Hi Wilbur,

thanks for the great piece of code!

Further to Thiago's question:
Since the result is seed-dependent, what would you say the best practice is? Shall I repeat the lassoregress call N times, pick the lambda with lowest MSE and run it one last time for that lmbda?

[John Riveros]

How should I interpret the significance of the estimators in this case where no significant test are available... ? I know it's a dummy question but I would like to know what you guys suggest about it.

[Colin Rowat]

What's the easiest way to compare train and test set MSE and R^2? With elasticnet, I can run lassogof after estimation. Is there an elasticregress equivalent?

[Colin Rowat]

How should I interpret the significance of the estimators in this case where no significant test are available... ? I know it's a dummy question but I would like to know what you guys suggest about it.

A simple answer would be: significance tests help prevent inclusion of 'irrelevant' variables, which can lead to overfitting. However, in a pure prediction task with a regularisation term, that term and cross-validation prevent overfitting.

Further, as Wilbur notes, the underlying distributions required for significance tests may not be available.