In discussing a fine grain evaluation of AL's behavior between configuration it occurred to use that there are probably several stochastic processes inside AL's various algorithms that likely make arbitrary choices based on some form of randomness. We should probably have some means for setting a universal seed for a run to verify that we can get the exact same learning behavior between runs. This issue is mainly to try and start a discussion to track down as many of these places as we can.
In discussing a fine grain evaluation of AL's behavior between configuration it occurred to use that there are probably several stochastic processes inside AL's various algorithms that likely make arbitrary choices based on some form of randomness. We should probably have some means for setting a universal seed for a run to verify that we can get the exact same learning behavior between runs. This issue is mainly to try and start a discussion to track down as many of these places as we can.