Make transformer encoder with high-frequency positional encoding and residual MLP decoder.

SS-Former Overview

MLP-Decoder of SS-Former

AI factory

Python 3.10.10  
tensorflow 2.12.0
tensorflow-addons 0.19.0
numpy 1.23.5
pandas 2.0.0
matplotlib 3.7.1
scikit-learn 1.2.2
tqdm 4.65.0

In this project, we use ETRI dataset.

We use 2018 and 2019 ETRI life log dataset.
There are 50 users, and total 5 activity classes.

0. If you want implement data merging, first you download 'dataset_2018.7z' files in ETRI.

1. And you some setting the 'Data_Merge_Processing.R' code.

For example)
1. Change setwd() function. 
2. Change the save path.
...

2. Implementation the R code.

1. Download our sample dataset. We provided user_06 and user_113 dataset.

2. Move to the file into data folder

data
├── user_6.csv
├── user_113.csv
└── ...

3. Implementation the 'data_preprocessing.ipynb'

If you want use whole variables or using another user etc, you will change among 'user_lst, var_lst, target_name'.

4. There are exists dataset.

data
├── user_6.csv
├── X_train.npy
├── X_valid.npy
├── X_test.npy
├── Y_train.npy
├── Y_valid.npy
├── Y_test.npy
└── ...

5. Implementation 'train.py'

# In python terminal
$(your path)> python train.py

• Performance of the proposed model for human activity classification for User_01

• The results with and without the use of high-frequency positional encoding for User_01

• Accuracy and F1-score of the proposed model for 20 randomly selected users