1 unstable release

0.1.0 Apr 11, 2022

#1541 in Algorithms

MIT license

19KB
180 lines

w_inter

A pair of solvers for the Weighted Interval Scheduling Problem.

Features

  • Zero external dependencies, although requires an allocator (not optional yet).
  • Flexible: anything implementing Ord + Add + Clone may be thought of as an interval bound or a weight type.
  • Efficient: running in O(n log n).
  • Fast: cache-aware, zero-reallocation APIs are available.

Simple Example

                   3───────────┐
                   └───────────┘
               8───────────────────┐
               └───────────────────┘
       5───────────┐   7───────────────┐
       └───────────┘   └───────────────┘
   3───────────────────────┐       4───────────┐
   └───────────────────────┘       └───────────┘
   2───┐       5───────┐   3───────────────┐
   └───┘       └───────┘   └───────────────┘
◀──0───1───2───3───4───5───6───7───8───9───10──11──▶
let intervals = vec![
  (0u8, 1u8,  2u8).into(), // (start, end, weight)
  (0u8, 6u8,  3u8).into(),
  (1u8, 4u8,  5u8).into(),
  (3u8, 5u8,  5u8).into(),
  (3u8, 8u8,  8u8).into(),
  (4u8, 7u8,  3u8).into(),
  (5u8, 9u8,  7u8).into(),
  (6u8, 10u8, 3u8).into(),
  (8u8, 11u8, 4u8).into()
];

let optimal = unsorted(&intervals);

assert_eq!(
  optimal,
  vec![
    (1u8, 4u8, 5u8).into(),
    (5u8, 9u8, 7u8).into()
  ]
);

Fast (Amortized Allocation) Example

// our goal is to allocate once and reuse the same buffers
// measure (or apply a guess) to avoid having to resize the vector.
let max_interval_count = problems.iter().map(|i| i.len()).max();

// we can say with certainty that the memo buffer
// will never need to be larger than the largest input size.
let mut memo = vec![0u8; max_interval_count];

// we don't know how big the solution set will be,
// but it can't be larger than the largest input size.
let mut soln = Vec::with_capacity(max_interval_count);

for intervals in problems {
  // perhaps we know our intervals to be *almost* sorted,
  // so we choose to use an algorithm tuned for this case.
  sort(&mut intervals);

  // we don't strictly need to, but we clear the old solution set,
  // so that only the optimal set from this run ends up in the buffer.
  soln.clear();

  sorted(
    &intervals,
    &mut memo,
    &mut soln
  );

  // we can now use the `soln` buffer before it's recycled
}

License: MIT

No runtime deps