ENH: warm-up period with exponential moving asgd and switch from sgd to asgd when empirical loss gets higher

npinto / asgd

Averaged Stochastic Gradient Descent Classifiers

41 stars 21 forks source link

Open npinto opened 13 years ago

npinto commented 13 years ago

The idea is to boost the performance by "disabling" the averaging until it gets useful. start with exp_moving_asgd_step_size=1e-2 ?