Adaptive Process-Guided Learning: An Application in Predicting Lake DO Concentrations

This repository is the official implementation of Adaptive Process-Guided Learning: An Application in Predicting Lake DO Concentrations submitted to ICDM-2024

Python environment

conda env create -f april.yml

Usage

(I) Data preprocessing

cd src/data
python3 main.py

(II) Train April

Tips: Before training April, we recommend downloading the pretrained discriminator model from this Google Drive link to achieve consistent results similar to ours. After downloading the discriminator models, please follow the steps in Section IV to generate data and Section VI for data preprocessing for April.

1. Pre-training stage

Shell script:

cd src/shell

./shell_pretrain_april.sh

Python script example:

cd src/train

python3 pretrain_main.py --model_type april --cluster_id 1 --model_index 1

Argument	Type	Default	Description	Choices
--model_type	str	'lstm'	The model to be used.	'lstm', 'pril', 'april', 'ea_lstm', 'transformer'
--gpu	int	0	GPU device ID to be used.
--model_index	int	1	Model index which maps to a random seed.	1,2,3
--cluster_id	int	1	Cluster Id	1,2,3,4

2. Fine tuning stage

Shell script

cd src/shell

./shell_finetune_april.sh

Python script example:

cd src/train

python3 pretrain_main.py --model_type april --cluster_id 1 --model_index 1

Argument	Type	Default	Description	Choices
--model_type	str	'lstm'	The model to be used.	'lstm', 'pril', 'april', 'ea_lstm', 'transformer'
--gpu	int	0	GPU device ID to be used.
--model_index	int	1	Model index which maps to a random seed.	1,2,3
--cluster_id	int	1	Cluster Id	1,2,3,4

(III) Train Pril

1. Pre-training stage

Shell script:

cd src/shell

./shell_pretrain_pril.sh

Python script example:

cd src/train

python3 pretrain_main.py --model_type pril --cluster_id 1 --model_index 1

2. Fine tuning stage

Shell script

cd src/shell

./shell_finetune_pril.sh

Python script example:

cd src/train

python3 finetune_main.py --model_type april --cluster_id 1 --model_index 1

(IV) Train discriminator model

Load existing discriminator models and generate data:

cd src/shell

./shell_use_dnn.sh

Retrain discriminator models and generate data:

cd src/shell

./shell_train_dnn.sh

example:

python3 train_DNN_all_lakes.py --model_index 1 --cluster_id 1 --TrainDNN 0

Argument	Type	Default	Description	Choices
--gpu	int	0	GPU device ID to be used.
--model_index	int	1	Model index which maps to a random seed.	1, 2, 3
--cluster_id	int	1	Cluster Id	1, 2, 3, 4
--TrainDNN	int	1	Set to 1 to retrain the discriminator model and generate data; set to 0 to load existing discriminator models and generate data.	0, 1

(V) Data preprocessing for April

cd src/data

python3 preprocess_extend.py

Directory Descriptions

Project Directory/
│
├── data/                   # Datasets storage
│
├── model/                  # Contains the April model and baseline architectures with related files
│
├── src/                    # Source code
│   ├── config/             # Configuration files (hyperparameters, environment variables, settings)
│   ├── data/               # Data preprocessing and loading functions
│   ├── draw_pics/          # Functions to generate visualizations
│   ├── evaluate/           # Model evaluation functions
│   ├── geo_map/            # Functions to generate geographical maps
│   ├── models/             # Model-related functions
│   ├── shell/              # Shell scripts
│   ├── train/              # Model training scripts
│   ├── utils/              # Utility functions
│
├── april.yml               # Environment configuration file for this project
│
└── README.md               # Project documentation

Contact

Should you have any questions regarding our paper or codes, please don't hesitate to reach out via email at chq29@pitt.edu or ruy59@pitt.edu.

Acknowledgment

Our code is developed based on GitHub - jdwillard19/MTL_lakes-Torch: Meta Transfer Learning for Lake Temperature Prediction.