Experiment: synthetic context generalization

kingjr / b2b

0 stars 0 forks source link

Open kingjr opened 5 years ago

kingjr commented 5 years ago

TODO

kingjr commented 5 years ago

JRR can outperform all models on both iid and validation set:

See code/example_validation.py

Note that I didnt manage to make it work with synthetic_v1 because of the covs

kingjr commented 5 years ago

This is simply a proof of existence. What remains to be done:

[ ] change code to use make_synthethic_v1 (I didnt manage because of the rolling cov, which I don't understand)
[ ] A systematic grid search over snr dim_x, dim_y etc