great paper thanks you

only can you clarify how it will work on big data
since your's set up
We used the same NN architecture across all methods: 2 layers of 64 hidden units with
ReLU non-linearities. We used the same learning rate, 1e
−3
, and the same batch size, 64,
for all methods. For all methods, training was stopped early if the validation loss did not
decrease for more than 200 epochs, until a maximum of 10000 epochs. If training was
stopped early, the final model was backtracked to the model with lowest validation loss.

would be limiting for big data
for example in comparison with
https://scikit-learn.org/stable/modules/generated/sklearn.svm.LinearSVR.html
??
lets say categorical data encoded to one hot then it will be
200000 parse binary features and 20 million rows?

do you know some code for these losses for simple regressors like LinearSVR?

is it possible to use use your code for linear regression?
I see you use pytorch , then it would be neural network without hidden layers ?

Could you please tell me where is the quantile_models? Or I need to define it by myself?