deprecate ArrayList and replace it with std::vector about mlpack

[rcurtin]

Commented by jcline3 on 13 Nov 40300065 02:13 UTC
Added a couple speed tests. Perhaps I used !ArrayList's initializers incorrectly, but it took significantly longer than std::vector to copy 1,000,000 ints.

[rcurtin]

Commented by rcurtin on 1 Jan 40303091 12:00 UTC
Ran the test and produced the following benchmarks:

:[ ryan @ greenstreet ]: ~ $ time ./testv

real    0m0.237s
user    0m0.136s
sys     0m0.104s
:[ ryan @ greenstreet ]: ~ $ time ./test

real    0m1.361s
user    0m1.288s
sys     0m0.076s

I think the next thing to see would be if ArrayList provides any functionality that std::vector does not, and investigate how to work around it, if it does. My concern here is our otrav mess.

[rcurtin]

Commented by jcline3 on 25 Mar 40365780 08:00 UTC
I found one interesting use of !ArrayList in fastlib/data/dataset.cc DatasetInfo::!ReadMatrix they build an !ArrayList, linearized, and then:

matrix->Own(linearized.ReleasePtr(), n_features, n_points);

  Elem *ReleasePtr() {
    DEBUG_MODIFY_OK(this);
    cap_ = -1;
    return ptr_;
  }

Setting cap_ to -1 is supposed to indicate the !ArrayList is now an alias, it seems.

  void Own(T *in_ptr, index_t in_length) {
    DEBUG_ONLY(AssertUninitialized_());

    ptr_ = in_ptr;
    length_ = in_length;
    should_free_ = true;
  }

This does prevent you from copying the entire matrix, however, there has to be a better way. Unfortunately, we cannot just do a straight swap of the !ArrayList with std::vector since then it attempts to free it twice. Fiddling with should_free_ may prevent this, though, it may also cause a memory leak. At any rate, this should be resolvable without having to copy of the data, even it it means rewriting or adding a couple functions. I am just concerned that we may run into more of this nonsense.

[rcurtin]

Commented by rcurtin on 9 Aug 40881622 04:53 UTC
(In #7) Let's wait until #6 and #24 are done before reorganizing the namespaces. After a lot of that cruft has been removed, it should make this a lot easier and more intuitive.

[rcurtin]

Commented by rcurtin on 14 Feb 41684598 16:20 UTC
(In #47) Because this is not compiled into MLPACK until sparse support is added, it is no longer blocking other tickets.

[rcurtin]

Commented by rcurtin on 4 Sep 41684615 03:04 UTC
(In #49) I am working on this, but because it is commented out of the build, it is not blocking #6 or #24.

[rcurtin]

Commented by rcurtin on 17 Jan 41684617 10:40 UTC
(In #40) This does not block #6 or #24.

[rcurtin]

Commented by rcurtin on 22 Dec 41684618 14:20 UTC
ArrayList is gone forever. Now, we can use more sensible objects instead. Hooray!