inconsistency of fixed lambda

[jonathan-taylor]

• Output value of fixedLassoInf is inconsistent for different likelihoods (i.e. gaussian does not return "sd" or "coef0")
• Gaussian seems to return one-sided p-values but Cox seems to be two-sided.
• The "coef" in the table is not the onestep estimator described here: http://arxiv.org/abs/1602.07358
• The "beta" argument of fixedLassoInf seems to be treated differently for logistic and gaussian. One allows an intercept, the other doesn't?

For numerical examples, see:

https://gist.github.com/jonathan-taylor/465d8fc2d5ba8b41d235b3ccd61dd217 https://gist.github.com/jonathan-taylor/c86b0e94b35edcf3b341af376c19a5ed