You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Experiment the implementation of moving average, and test if its smoothing effect helps training performance.
Dataset: IBM stock price dataset, 362 samples
Set decay = 0.9
Result:
Calculate the moving average by hand directly:
Calculate the exponential moving average by tensorflow EMA function:
The decay rate is set by min(decay, (1 + num_update)/(10 + num_update)), which means actual decay rate increase from 1/10 to decay rate 0.9.
About
Experiment the implementation of moving average, and test if its smoothing effect helps training performance.