COMP4560_stokes_ml_project

A repository that we are going to use to keep track of project evolution, notes, ideas, etc.

I am going to use this repository as a sort of research diary with my own notes on the project.

Geoid Problem

Datasets

Datasets used for this problem are located in the folder Data/Geoid/new_results_1k_zero.

Files

Files related to this problem include:

Geoid_systematic_training.ipynb
Geoid_visualisation.ipynb
ModelList.txt

Geoid_systematic_training.ipynb is used for systematically training the models with their hyperparameters in the file ModelList.txt and saving them in the folder 1D_result.

Geoid_visualisation.ipynb is used for visualisation given a specific model's path.

NOTE: pip install cartopy and pip install pyshtools need to be run before using Geoid_visualisation.ipynb for visualisation.

Mantle Convection Problem

Supported by HPC

This problem is mainly researched with the help of a HPC system called Gadi. Files in the repository with a suffix of .sh are shell scripts submitted to Gadi in order to run the python script on Gadi.

Datasets

Datasets used for this problem can be found in the following URLs (accessible for ANU students only):

• Limited Dataset
• Larger Dataset
• Interpolated Dataset

Interpolated dataset is generated from the Larger dataset using interpolation.py and interpolation_job.sh.

Convolutional Autoencoders (ConvAE)

Convolutional Autoencoder is used to compress the data before feeding the data into a predicting model.

Files related include:

ConvAE_training.py
ConvAE_testing.py
ConvAE_training_job.sh
ConvAE_testing_job.sh
ConvAE_visualisation.ipynb

Training and testing results are stored in the folder 2D_ConvAE_results

Fully Connected Neural Network (FNN)

Fully Connected Neural Network is used to predict the temperature field at the next time step.

Files related include:

FNN_training.py
FNN_testing.py
FNN_training_job.sh
FNN_testing_job.sh
FNN_visualisation.ipynb

Training and testing results are stored in the folder 2D_FNN_results.

Long Short-Term Memory (LSTM)

Long Short-Term Memory is used to predict rest of the temperature fields as a sequence given the first 50 temperature field in a simulation.

Files related include:

LSTM_training.py
LSTM_testing.py
LSTM_training_job.sh
LSTM_testing_job.sh
LSTM_visualisation.ipynb

Training and testing results are stored in the folder 2D_LSTM_results.