# 110th place solution 

Acknowledgement
I want to thank my teammates
[@mahdiseddigh](https://www.kaggle.com/mahdiseddigh)
,
[@shanzhong8](https://www.kaggle.com/shanzhong8)
,
[@guoooooooss](https://www.kaggle.com/guoooooooss)
and
[@leonshangguan](https://www.kaggle.com/leonshangguan)
. You all put in your best effort to manage your time for this tough competition. Even though we didn’t perform well last month, at least we got the chance to review our college physics course knowledge.
Solution
We observed that different behaviors lead to different results depending on the phase. This suggests that carefully manipulating the signal during a significant phase could be highly beneficial, as the phase plays a critical role in determining the outcome.
Quaternion error:
![quaternion error](assets/7591ccd24ef379645e1e6d22.png)
Gravity accel error:
![gravity accel error](assets/e51c3e5376d714347ac1caa3.png)
Then it turns out that we find a good library to handle this type of signal, using ATAR algorithm in
[spkit](https://spkit.github.io/modules/generated/spkit.eeg.ATAR.html#spkit.eeg.ATAR)
to process the signal in each phase, and boost cv and lb.
Here are the result using different parameter in ATAR on acceleration:
![](assets/fc57763e09ebca323b10e38e.png)
We also try to use
[Faiss](https://github.com/facebookresearch/faiss)
to retrieve similar signal and use them in the inference, just miss to use subject which is really important in this competition.
Will fill up more later.