Zhuyi Xue

In general, I thought it would be S: sequence length B: batch size N: embedding size or hidden layer size (i.e. number of RNN cells in a hidden layer)

Hi @spro, I am having the same question, could you please point out which sentence mentioned the $g$? Thanks! The paper is at https://arxiv.org/pdf/1508.04025.pdf

@adamklec, do you know where that non-standard attent is from? Thanks.

@jameslamb , gotcha, thank you for the tip.

Is https://github.com/microsoft/LightGBM/blob/master/include/LightGBM/utils/json11.h expected to be the same as https://raw.githubusercontent.com/dropbox/json11/master/json11.hpp ? the example on https://github.com/dropbox/json11 doesn't seem to work with the json11 in lightgbm...

@jameslamb , I've reimplemented `LGBM_BoosterGetConfig` to return parameters as a json string, please let me know what you think (I'm still relatively new to c++).

>please don't make any changes to the organization until another maintainer like @StrikerRUS, @shiyu1994, or @tongwu-msft comments. Hey @jameslamb , should I update now or wait till more comments come...

Yes, I specified `--rm -t -i`, and then `py.test` interactively inside the docker container.

Just to add that this happens when the test is run on a jenkins server, as well if the output is sent to the browser.

@alexcoca , do you have insight (or intuition) into why `tree_path_dependent` is a valid way of estimating `E[f(x)|x_S]`, esp. when S is not the empty set? I haven't found any...