top-k

Keep the top-k highest-priority items in O(k) space. Zero runtime dependencies.

Internally, TopK maintains a min-heap of (priority, unique_tiebreaker, item) entries via the standard-library heapq. The heap root is the lowest stored priority — the eviction candidate when the structure is full. Each insertion compares the new item's priority against that root and replaces it only when the candidate is strictly better, so memory stays proportional to k regardless of how many items you consider.

Table of Contents

• Key Features
• Installation
• Quick Start
• API
• Use Case: Streaming Top-K
• Priority Mutation
• Development
• License

Key Features

• High-level abstraction for "keep the best k items". The priority is a callable on your item type. Return -score to keep the k least risky items instead.
• O(k) space. Only the current top-k is stored. Additional candidates are either rejected or swapped with the current lowest-priority entry.
• Generic and fully typed. TopK[T] stores your items as-is, including third-party types you do not control. You pass a priority callable (Callable[[T], int | float]) instead of wrapping those objects in (score, item) tuples. Type checkers can still validate that every stored item is a T. Ships with a py.typed marker (PEP 561). Works out of the box with mypy, pyright, and other type checkers.
• Items need not be comparable. Many third-party types have no __lt__. A monotonic unique tiebreaker is stored alongside each priority so heapq never falls through to T.__lt__ — you do not have to wrap the item just to make it heap-safe.
• Zero runtime dependencies. The package uses only the Python standard library (heapq).
• Tested on Python 3.12 through 3.14.

Installation

pip install top-k

Or with uv:

uv add top-k

Quick Start

from dataclasses import dataclass

from top_k import TopK


@dataclass(slots=True)
class UrlRisk:
    url: str
    score: float
    source: str
    confidence: float


top_k = TopK[UrlRisk](3, priority=lambda u: u.score)

for item in (
    UrlRisk("https://phish.example/login", 95.0, "urlhaus", 0.9),
    UrlRisk("https://malware.example/payload", 88.0, "sandbox", 0.8),
    UrlRisk("https://steal.example/oauth", 72.0, "urlhaus", 0.7),
    UrlRisk("https://ads.example/banner", 40.0, "sandbox", 0.6),
    UrlRisk("https://cdn.example/static", 8.0, "sandbox", 0.9),
):
    top_k.add(item)

assert len(top_k) == 3
assert (
    top_k.smallest is not None and top_k.smallest.url == "https://steal.example/oauth"
)

smallest is the lowest priority among the items currently stored (the heap root), not the highest-scoring URL. items yields stored objects in heap order, not sorted by priority.

API

Use Case: Streaming Top-K

heapq.nlargest is the right tool when every candidate is already in memory. TopK is for an online stream — a log file, a socket, or an object-store listing — where lines arrive one at a time and you must not retain the full collection.

from collections.abc import AsyncIterator
from dataclasses import dataclass

from top_k import TopK


@dataclass(slots=True)
class LogLine:
    severity: int
    message: str


async def iter_log_lines(path: str) -> AsyncIterator[LogLine]: ...


async def top_errors(path: str) -> TopK[LogLine]:
    chosen = TopK[LogLine](10, priority=lambda line: line.severity)
    async for line in iter_log_lines(path):
        chosen.add(line)
    # At most 10 lines retained, even if the stream has millions.
    return chosen

Peak memory is O(k), independent of stream length.

Priority Mutation

items returns the same item objects held internally. Changing an item's priority (or any field that priority reads) breaks heap correctness. Treat those items as read-only.

Stored priorities are computed at add time; this structure does not re-sort if you change an item later. The unique tiebreaker prevents heapq from comparing the raw items, so in-place changes will not raise TypeError — they still corrupt the heap.

Development

git clone https://github.com/ori88c-python-packages/top-k.git
cd top-k
uv sync

# Run tests
uv run pytest

# Lint and format
uv run ruff check .
uv run ruff format .

# Type check
uv run mypy src

The same three commands run on git commit via pre-commit. Format in the hook is ruff format --check ., so an unformatted tree fails the commit instead of rewriting it. After uv sync, install the hook once:

uv run pre-commit install

License

Apache 2.0