[mlpack-git] [mlpack/mlpack] Universal B tree implementation (#746)

[lozhnikov]

> +  // of hyperrectangles is restricted.
> +  size_t numCorners = 0;
> +  for (size_t pos = numEqualBits + 1; pos < order * tmpHiAddress.n_elem; pos++)
> +  {
> +    size_t row = pos / order;
> +    size_t bit = order - 1 - pos % order;
> +
> +    // This hyperrectangle is not contained entirely in the bound.
> +    // So, the number of hyperrectangles should be increased.
> +    if (tmpHiAddress[row] & ((AddressElemType) 1 << bit))
> +      numCorners++;
> +
> +    // We ran out of the limit of hyperrectangles. In that case we enlare
> +    // the last hyperrectangle.
> +    if (numCorners >= maxNumBounds / 2)
> +      tmpHiAddress[row] |= ((AddressElemType) 1 << bit);

yeah, I think if `maxNumBounds` is odd then we end up with `maxNumBounds - 1`. I think we can replace
`if (numCorners >= maxNumBounds / 2)`
by
`if (numCorners >= maxNumBounds - numBounds)` in `InitLowerBound()`
In that case we should document that `InitLowerBound()` should be called after
`InitHighBound()`. But in that case the functions are non-symmetric.

Another way is to document that `maxNumBounds` should be even.

And another way is to make an extra pass and calculate `numCorners` for the lower bound and for the high bound. What do you think about these variants?

-- 
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/mlpack/mlpack/pull/746/files/f17843fc8e8ef2b1c4b04d572c521575f20a1f3c#r76321604
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.cc.gatech.edu/pipermail/mlpack-git/attachments/20160825/f8d2a2ff/attachment-0001.html>