[mlpack-git] [mlpack/mlpack] Heaps for mlpack! (#732)

[MarcosPividori]

Hi @rcurtin , I have run some tests with *mlpack_knn* and the *"Isolet"* dataset. It looks like it slightly faster when k is small, and the difference is greater for bigger values of k.
Maybe, we should benchmark with bigger datasets like *"tinyImages100k"*. Also, it would be interesting to see the effect in other methods, like: *mlpack_cf*, *mlpack_lsh*, *mlpack_fastmks*, *mlpack_krann*, etc.


# Naive:
 + **No-heap:**
  - k=3: total_time: 32.790772s
  - k=10: total_time: 32.946072s
  - k=20: total_time: 32.869260s
  - k=100: total_time: 33.801901s
  - k=1000: total_time: 52.723884s

 + **Heap:**
  - k=3: total_time: 31.996061s
  - k=10: total_time: 32.623913s
  - k=20: total_time: 32.756599s
  - k=100: total_time: 33.620033s
  - k=1000: total_time: 44.275435s 

# Single-tree search:
 + **No-heap:**
  - k=3: total_time: 19.419358s
  - k=10: total_time: 21.696434s
  - k=20: total_time: 23.501047s
  - k=100: total_time: 29.048157s
  - k=1000: total_time: 57.329319s

 + **Heap:**
  - k=3: total_time: 19.386155s
  - k=10: total_time: 21.643011s
  - k=20: total_time: 23.270618s
  - k=100: total_time: 29.031573s
  - k=1000: total_time: 50.519661s

# Dual-tree search:
 + **No-heap:**
  - k=3: total_time: 15.207993s
  - k=10: total_time: 17.129599s
  - k=20: total_time: 18.325870s
  - k=100: total_time: 22.422410s
  - k=1000: total_time: 48.341919s

 + **Heap:**
  - k=3: total_time: 15.203970s
  - k=10: total_time: 16.998384s
  - k=20: total_time: 18.240207s
  - k=100: total_time: 22.401873s
  - k=1000: total_time: 41.206338s

---
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/732#issuecomment-234604657
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.cc.gatech.edu/pipermail/mlpack-git/attachments/20160722/8d316e90/attachment-0001.html>