Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,36 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
tags:
|
| 4 |
+
- GUI
|
| 5 |
+
- Agents
|
| 6 |
+
pretty_name: Android Control Test Set
|
| 7 |
+
---
|
| 8 |
+
# Android Control Test Set
|
| 9 |
+
|
| 10 |
+
## Notes
|
| 11 |
+
1. **NOTE** i uploaded it so it easier for running evaluation on it
|
| 12 |
+
2. I am not responsible if you use it in training :<
|
| 13 |
+
3. I assume most models train on train-set/val-set but not test-set
|
| 14 |
+
4.
|
| 15 |
+
I assume that the ids here are episode_id
|
| 16 |
+
https://console.cloud.google.com/storage/browser/_details/gresearch/android_control/splits.json , so I download data using smolagents
|
| 17 |
+
https://huggingface.co/datasets/smolagents/android-control and iterate and filter ids to have only the real test set
|
| 18 |
+
|
| 19 |
+
## IMPORTANT
|
| 20 |
+
(notice smolagents dataset repo contain some part of partitioning but episode_id in it are different from the main source of data)
|
| 21 |
+
|
| 22 |
+
## Repo Structure
|
| 23 |
+
there small parquet files for downlading samples and explore it
|
| 24 |
+
then there's a huge zip-file (I figured it easier to download if I am going to eval on whole set inside kaggle or colab or other providers)
|
| 25 |
+
|
| 26 |
+
Thanks
|
| 27 |
+
|
| 28 |
+
## Citations
|
| 29 |
+
```
|
| 30 |
+
@article{li2024effects,
|
| 31 |
+
title={On the Effects of Data Scale on Computer Control Agents},
|
| 32 |
+
author={Li, Wei and Bishop, William and Li, Alice and Rawles, Chris and Campbell-Ajala, Folawiyo and Tyamagundlu, Divya and Riva, Oriana},
|
| 33 |
+
journal={arXiv preprint arXiv:2406.03679},
|
| 34 |
+
year={2024}
|
| 35 |
+
}
|
| 36 |
+
```
|