@jscuds @vsavram @mcembalest @mecunha I created a config.py file with the default hyper parameters for our models. A few things to note:
don't change the config!!! change the values in your python notebook.
take a look at NLMdemo or LUNAdemo to see how the config gets called and used: from config import *
max, elaine, and I will notify everyone if something in the config changes
the default models run 100 iterations. currently NLM and feedforward take about 1 second to train and LUNA takes 2 minutes.
if you want to run the real deal model, set opt_params["max_iteration"] = real_max_iteration causing the model to train for 5000 iterations. This takes LUNA about 3 hours to run, and NLM like 8 minutes.
if you want to fit a feed forward object, use the nlm_architecture
@jscuds @vsavram @mcembalest @mecunha I created a config.py file with the default hyper parameters for our models. A few things to note:
from config import *opt_params["max_iteration"] = real_max_iterationcausing the model to train for 5000 iterations. This takes LUNA about 3 hours to run, and NLM like 8 minutes.nlm_architecturelet me know if you have any questions!!