Jaak Simm

Support for additional heads

1

I'd be interested to create environments that have add extra data for auxiliary training tasks available (e.g., reward prediction). Is that already possible, thx! I don't mind rewriting some methods...

Seed is taken as input but not passed to Python API in generate_folds call.

Thank you for the code. I did a quick check with revin to see if it is causal along the sequence length axis, where by "causal" I mean that only...

Good speed-ups will be gained by using FastIDF.

Currently distributed matrix product is already done, but uses `Vector{RemoteRef}`. Wrap it into a type and add support for `*`, `At_mul_B` etc. Good for readable and manageable code.

For supporting factor inputs for Z in "dea.env.robust.R" we need to 1. remove check of `Z` being matrix 2. drop using `Z1` and instead use `model.matrix` (or some predict/fit function...

Current version uses distance to for linking features to the efficiency (beta_hat, beta_hat_at etc). Provide an option to use theta (efficiency).

dea.env.robust needs to documentation for outputs - beta_hat_hat_star - delta_ci_kneip_{low,high}

Add Tone's costmin to dea, dea.robust, dea.env.robust.

Why is noise scaled by Ntrain in RMSProp

7

In `SGLD_RMSprop.m` the noise is scaled by `opts.N` which is set to Ntrain in DNN experiments: https://github.com/ChunyuanLI/pSGLD/blob/master/pSGLD_DNN/algorithms/SGLD_RMSprop.m#L51 Why is this the case? In the paper (https://arxiv.org/pdf/1512.07666v1.pdf) there is no such...