Skip to content
This repository has been archived by the owner on Sep 23, 2019. It is now read-only.
André Setti edited this page Sep 3, 2018 · 6 revisions

Welcome to the Stardew Valley Fishing Bot (SVFB) wiki!

This project is for people like me, who can't fish for shit on this game, so we can have a chance totaling it! It's still not complete, we need help collecting data from the game to finish the machine learning process.

How we capture data

Our option to capture what is happening on the screen is using OpenCV, an open-source library for computer vision which supports python 3.6. There may be better option to capture the game screen data, but as this was my first big project I didn't think a lot about this. Another option we think is possible is to access the game data in memory directly, but we don't know how to do that and, sincerely, we are already at 60FPS so it won't make much difference anyway.

To capture your mouse-clicks or key presses we use pywin32, another open-source library, which allows python to integrate services with windows. For this reason, this bot is not compatible with Linux or Mac (we didn't even test it, but it should not work anyway).

What we send to our server

We only capture your in-game fishing screen and your mouse-clicks (or key-presses) when you are fishing. And that is only what we use for the bot training.

As we are poor students and can't afford fancy servers with a lot of space, we try to only use essential data for the bot. We don't send anything more than those in-game screen captures.

Also, you can see and manage the data sent to the server in the website if you want to delete them. If some miracle happens and we stop being poor we will enable users to download their data directly from our server too.

What we do with the data

We plan to automatically train a new Convolutional Neural-Network (CNN) every time the server receives a certain amount of new data, or after a certain period of time passes. This is to ensure there will be a new Model to download which should be better every new release.

We also plan to allow users to train their own CNN with their offline data if they want (not available yet).