hhchen1105/Generating-Personalized-Trend-Line-Based-on-Few-Labelings-from-One-Individual

Generating-Personalized-Trend-Line-Based-on-Few-Labelings-from-One-Individual

This repo contains the code to reproduce the result in the paper:

Petrel: Personalized Trend Line Estimation with Limited Labels from One Individual published in PAKDD 2023.

Petrel generates customized trend lines that consider user preferences and input time series shapes. Petrel obtains users' preferred trends by asking users to draw trend lines on limited sample trends.

Setup

Tested under Python 3.8.13 in Ubuntu.

Install the required packages by

conda env create -f environment.yaml
conda activate env_tykuo

File Description

Data Folder Link: https://drive.google.com/file/d/1bc2A4OKITtgTyA1ZDQAnTYl2NLcE7nDC/view?usp=sharing

• A4Benchmark: YahooS5 A4 Dataset csv file
• A4Benchmark_SimulateUser: Generate simulate User
• Done_s2_img_user: The pictures that model predict with differnt user
• Done_user: The user data json file and pictures that user draw.
• mixer_multiple_full: Generate simulate user by GenerateSimulateUser.py
• trend: original time series dataset which include value, L1 Trend Filtering, HP Filtering, STL trend.

Execution

Folder Setting

project
│   README.md
│   evvironment.yaml    
│   pretraincnn_model
│   pretrainfc_model
│   pretraintransformer_model
│   pretrainlstm_model
│   OurMethod_model
│
│
└───A4Benchmark_SimulateUser
│   │   A4Benchmark-TS1.csv
│   │   A4Benchmark-TS2.csv
│
└───A4Benchmark
│   │   A4Benchmark-TS1.csv
│   │   A4Benchmark-TS2.csv
│   
└-───Trend
│   │   file1.json
│   │   file2.json
│   │   file3.json 
│
└───mixer_multiple_full     
│   
└───user29(User who you want to check)
│   │   file1.pdf
│   │   file2.pdf
│   │   user29.json(User draw data) 
│
│
└───s2_img_user29(User who you want to check)
    │   file1.pdf
    │   file2.pdf
    │   user29.json(User draw data) 

Initialize folder

To store model output pictures

mkdir mixer_multiple_full
mkdir s2_img_user29
cd s2_img_user29
mkdir 1_v2 8_v2 12_v2 20_v2 21_v2 66 74 81 88 91

GenerateSimulateUser

You need to run this code to generate simulate users' data before you run Pretrain model code and OurMethod code.

python GenerateSimulateUser.py

CNN

python CNN.py --epoch 3 --lr 0.01 --batch 10 --user 29

LSTM

python LSTM.py --epoch 3 --lr 0.01 --batch 10 --user 29

Transformer

python Transformer.py --epoch 3 --lr 0.01 --batch 10 --user 29

PretrainCNN

Train

python PretrainCNN_train.py --epoch 1 --lr 0.01 --batch 1000 

Finetune and Test

python PretrainCNN_finetune_test.py --epoch 3 --lr 0.01 --batch 10 --user 29

PretrainLSTM

Train

python PretrainLSTM_train.py --epoch 1 --lr 0.01 --batch 1000

Finetune and Test

python PretrainLSTM_finetune_test.py --epoch 3 --lr 0.01 --batch 10 --user 29

Pretrain Fully Connected

Train

python PretrainFullyConnected_train.py --epoch 1 --lr 0.01 --batch 1000

Finetune and Test

python PretrainFullyConnected_finetune_test.py --epoch 3 --lr 0.01 --batch 10 --user 29

Pretrain Transformer

Train

python PretrainTransformer_train.py --epoch 1 --lr 0.01 --batch 1000

Finetune and Test

python PretrainTransformer_finetune_test.py --epoch 3 --lr 0.01 --batch 10 --user 29

OurMethod

Train

python OurMethod_train.py --epoch 1 --lr 0.01 --batch 1000

Test

python OurMethod_test.py

L1 Trend Filtering & HP Filtering & STL

python l1hpstl.py

Citation

Please cite our work if you find Petrel useful in your research.

@inproceedings{
    kuo2023petrel,
    title={Petrel: Personalized Trend Line Estimation with Limited Labels from One Individual},
    author={Tong-Yi Kuo and Hung-Hsuan Chen},
    booktitle={The 27th Pacific-Asia Conference on Knowledge Discovery and Data Mining},
    year={2023}
}