37 lines
1.2 KiB
Markdown
37 lines
1.2 KiB
Markdown
title: Top Packages Algorithm
|
|
|
|
## Package Score
|
|
|
|
Each package is given a `score`, which is used when ordering them in the
|
|
"Top Games/Tools/Asset Packs" lists. The intention of this feature is
|
|
to make it easier for new users to find good packages.
|
|
|
|
A package's score is equal to a rolling average of recent downloads,
|
|
plus the sum of the score given by reviews.
|
|
|
|
A review score is 100 if positive, -100 if negative.
|
|
|
|
```c
|
|
reviews_sum = sum(100 * (positive ? 1 : -1));
|
|
score = avg_downloads + reviews_sum;
|
|
```
|
|
|
|
## Pseudo rolling average of downloads
|
|
|
|
Each package adds 1 to `avg_downloads` for each unique download,
|
|
and then loses 5% (=1/20) of the value every day.
|
|
|
|
This is called a [Frecency](https://en.wikipedia.org/wiki/Frecency) heuristic,
|
|
a measure which combines both frequency and recency.
|
|
|
|
"Unique download" is counted per IP per package.
|
|
Downloading an update won't increase the download count if it's already been
|
|
downloaded from that IP.
|
|
|
|
## Transparency and Feedback
|
|
|
|
You can see all scores using the [scores REST API](/api/scores/), or by
|
|
using the [Prometheus metrics](/help/metrics/) endpoint.
|
|
|
|
Consider [suggesting improvements](https://github.com/minetest/contentdb/issues/new?assignees=&labels=Policy&template=policy.md&title=).
|