Repeated parameter values

[ameenetemady]

Hi,

I see Spearmint sometimes provides the exact same parameter value combination (on which my objective function is already evaluated in a previous iteration). Is that expected? Note: my parameters were all of type "INT".

Thanks,
Ameen.

[satarupaguha11]

I am facing the same issue. After a few iterations, all parameter values become 1.0 and continue to remain so until I manually abort the program.
Can anybody help with this?

[JasperSnoek]

It sounds like the model is having trouble fitting the function values. In
this implementation, we assumed that the observations were a relatively
smooth function of the integers. If that isn't the case, then you can have
strange behavior (consider fitting a polynomial to e.g. a step function -
it will wildly overshoot on either side of the step in order to maintain
smoothness). This can cause huge bumps in the estimated function that will
correspond to a single integer value having unrealistically high
probability of being the best.

On Tue, Dec 15, 2015 at 7:20 AM, Satarupa Guha [email protected]
wrote:

I am facing the same issue. After a few iterations, all parameter values
become 1.0 and continue to remain so until I manually abort the program.
Can anybody help with this?

—
Reply to this email directly or view it on GitHub
#46 (comment)
.

[wuzhou-zhang]

I am facing the same issue. My parameter is double. And the repeated parameter value is not the best parameter value.

[gwl2108]

I'm also getting this issue. Is there a suggested workaround? I tried playing with my objective function but it hasn't helped