If I understand correctly the standard optimization method model.optimize(method='TNC') returns the negative log likelihood while SGD model.stochastic_optimize(snps_per_minibatch=1000, num_iters=10, svrg_epoch=3) returns the negative log likelihood which is a bit confusing unless I am misunderstanding the values that are returned under the log_likelihood parameter.
Sorry for not responding to this for so long. You're right, the log_likelihood entry in the result returned by stochastic_optimize is misnamed. I renamed that field to cross_entropy instead.
If I understand correctly the standard optimization method
model.optimize(method='TNC')
returns the negative log likelihood while SGDmodel.stochastic_optimize(snps_per_minibatch=1000, num_iters=10, svrg_epoch=3)
returns the negative log likelihood which is a bit confusing unless I am misunderstanding the values that are returned under the log_likelihood parameter.