Of these four libraries, which are you most likely to use?

I'm trying to pick out my next hackery project. It'll likely be one of the following:

• A sparse radix trie Implementation with extremely fast set operations
• A really good soft heap implementation
• A bloomier filter implementation
• A collection of small financial algorithms, such as deriving total returns given a set of dividends and minimal information about them.

But I can't choose. So I thought I'd put my fate in the hands of my peers. Which of those four would you find most useful? Most interesting to work on? Which do you think is the most needed?

I didn't know what a bloomier (maybe Bloom?) filter is until reading your question. Sounds cool and useful.