BalancingLearner: add a "cycle" strategy, sampling the learners one by one #188
Conversation
basnijholt
force-pushed
the
cycle_strategy
branch
2 times, most recently
from
1ab7a88
to
396fa52
Compare
basnijholt
force-pushed
the
cycle_strategy
branch
from
396fa52
to
858ca89
Compare
| @@ -176,6 +181,20 @@ def _ask_and_tell_based_on_npoints(self, n): | |||
| points, loss_improvements = map(list, zip(*selected)) | |||
| return points, loss_improvements | |||
|
|
|||
| def _ask_and_tell_based_on_cycle(self, n): | |||
| if not hasattr(self, '_cycle'): | |||
There was a problem hiding this comment.
This could also just be set in the __init__.
|
I think this is rather an indication of If this is a preferable implementation, we should rather remove the other strategy. However it'd make most sense if |
|
But |
|
I see, that's different indeed. Is that a behavior we would want to support? It is also different from essentially everything else in adaptive: it tells the learner what to do and now what goal to reach. |
basnijholt
force-pushed
the
cycle_strategy
branch
from
858ca89
to
e6afb71
Compare
| @@ -173,6 +178,20 @@ def _ask_and_tell_based_on_npoints(self, n): | |||
| points, loss_improvements = map(list, zip(*selected)) | |||
| return points, loss_improvements | |||
|
|
|||
| def _ask_and_tell_based_on_cycle(self, n): | |||
| if not hasattr(self, "_cycle"): | |||
| self._cycle = itertools.cycle(range(len(self.learners))) | |||
There was a problem hiding this comment.
do this in __init__ as I commented above.
There was a problem hiding this comment.
But you wrote to do in the strategy property setter?
basnijholt
force-pushed
the
cycle_strategy
branch
from
e6afb71
to
d25f3d1
Compare
|
@jbweston I think I've done what you suggested (up to #188 (comment)). Merge if you like. |
| @@ -107,10 +110,13 @@ def strategy(self, strategy): | |||
| self._ask_and_tell = self._ask_and_tell_based_on_loss | |||
| elif strategy == "npoints": | |||
| self._ask_and_tell = self._ask_and_tell_based_on_npoints | |||
| elif strategy == "cycle": | |||
| self._ask_and_tell = self._ask_and_tell_based_on_cycle | |||
| self._cycle = itertools.cycle(range(len(self.learners))) | |||
There was a problem hiding this comment.
Now the cycle will be reset every time the strategy is set dynamically. I'm not sure what the best thing to do here is.
There was a problem hiding this comment.
That might be the intention? We could also just put in in __init__...
This is useful for example for the
AverageLearner1Dand2Dwhere sampling onnpointswon't sample all learners equally, because that's not hownpointsis defined there.