Dataset not appearing after recording

[sDionys]

Hello,

I found an issue with the launcher.

After recording gameplay in Teach mode, the dataset should appear in the Datasets section so it can be used for training. However, the dataset never shows up in the launcher, and because of that I cannot start training.

What I tested:

Recorded gameplay normally
Tried multiple different games
Reinstalled the launcher completely
Restarted the application and PC

Result:

Recording seems to save successfully (Recording saved)
No dataset appears in the Datasets panel
Training cannot be started because no dataset is detected

This issue happens with all games, not just one specific game.

Version: v0.2.1 (Stable)

[Toanl31]

Seems to be a fix here. I've asked the user for what they mean with the command line but no reply #57 (comment)

I'd assume you need to do the training part with command line which is tedious and not very intuitive as many users are not very knowledgeable with cli hence most use the .exe version.

[ruslanmv]

Thank you for your reply and for sharing your feedback.

We’ll look into this and work on improving the experience, especially for users who prefer using the launcher rather than the command line.

[ruslanmv]

Thanks for the detailed repro — this turned out to be a path mismatch on the recording side.

Root cause: the Rust start_recording handler in src-tauri/src/main.rs was spawning 1-collect_data.py with no --out argument, so the script fell back to its default (data/raw). That path is outside the datasets/<game_id>/<dataset_name>/ layout that:

• SessionManager.finalize_recording snapshots for new .npy files,
• the /modelhub/datasets endpoint scans (via _scan_datasets_fs),
• and the start-recording preflight reserves at main.rs:4863.

End result: recording "succeeded", the files landed in the wrong tree, and the Train tab kept showing "No datasets yet".

Fix on branch claude/fix-open-issues-7LDKz:

1. start_recording now passes --out <local_data_root>/datasets/<gid>/<name> so the .npy files are written exactly where the session manager and the dataset listing both look.
2. _scan_datasets_fs now also surfaces a synthetic "Unsorted recordings" entry when loose .npy files exist at the root of datasets/<gid>/, so anyone with files from the old layout still sees something in the Train tab and can recover.

If you can rebuild from that branch, the dataset should appear immediately after you stop recording — no relaunch needed.

---

Generated by Claude Code

[Toanl31]

@ruslanmv I haven't tested anything said. I just wanted to take this time to appreciate the work you are putting in. I want to eventually start using this and use it as a hobby to play games. We're all mostly old now, I used to watch MMOHUT videos (I know for sure anyone reading my comment and plays MMORPG knows about MMOHUT or MMORPG.COM and all that stuff) so I don't really have the time to play anymore, but with this Bot I might have.

Thank you for your work.

[ruslanmv]

@Toanl31 Thank you so much for your kind words, I really appreciate it.

I am in the same boat. I have a family, I work a lot, and I also do not have much time to play anymore. But video games, especially MMORPGs, are still one of my passions. That is one of the reasons I started developing this project: so I could relax, stay connected with the games I love, and enjoy them in a different way despite having limited time.

Comments like yours are very motivating and remind me why this work is worth doing. I still have a lot to improve and fix, but I am happy that this project may be useful for people who share the same feeling.

Thank you again for your support.

[sDionys]

@ruslanmv

Thank you for addressing the issue and implementing these improvements.

I previously used the console workflow for dataset training, but the new interface makes the process significantly more user-friendly and efficient.

In my opinion, this project has strong potential for the future of MMORPG automation, especially because it avoids direct game process injection and instead focuses on adapting to real player behavior patterns.