Eric Liu
Eric Liu
BLOCKED ON: #274 Currently I have things like: ComputeOptimalPointToSampleWithRandomStarts and ComputeOptimalPointToSampleViaMultistartGradientDescent The only difference here is that the second one lets you specify the starting_points for MGD and the first...
See comments from @ypwais on this review: https://github.com/sc932/MOE/pull/268 Clearing out ExpectedImprovementEvaluator/State ctor arguments from ComputeOptimalPointsToSample (and friends) would also go a long way toward cleaning this mess up.
Currently GradientDescentParameters and NewtonParameters lets you specify a "num_multistarts" field. But for 'dumb' search, this is specified as a plain int. This doesn't make a lot of sense. Remove num_multistarts...
From my comment on #291 worth double checking how exactly colander deals with the `missing=` and `default=` fields in schemas. I think they're copied by reference so that somehow changing...
From my comment on #291 lots of colander examples appear to instantiate schemas immediately after definition so that they're only instantiated once. We do a mix of A) instantiate as...
RELATED TO: (GH-192) inserting rows into Cholesky It's probably easiest to prototype this in Python and then implement in C++. Note that this is trivial if you're removing the _last_...
Right now gradient descent doesn't offer much introspection into what it did & whether it succeeded: 1) indicate whether convergence occurred (e.g., did the gradient drop below tolerance?) 2) indicate...
One of the bugs found in #82 was: ``` DEFAULT_GRADIENT_DESCENT_MULTISTARTS = 10000 DEFAULT_OPTIMIZATION_MULTISTARTS = 10000 ``` Those values should've been more like a few hundred. If we reported (in http...
This follows a comment I made in #241 which was out of scope for the that pull request. This issue is also touched on in #34. The current proposition is:...
Since moe expects numpy arrays as input and gives numpy arrays as output, it'd be nice if colander supported this directly. Currently (as a hack), I (eliu) call numpy.array() on...