Dataset

The dataset can be found at https://www.kaggle.com/datasets/rainmaker29/sdss-images

Environment

To set up environment, please install the libraries from libraries_installed.txt using conda

Reproduce results

To reproduce the results, please run the following files:

Numeric modelling

1. Data Preprocessing : .\Numeric data modelling\Testing_Normal_distribution_features.ipynb
2. Traditional models on numeric data : .\Numeric data modelling\Traditional ml algorthims.ipynb

Image Modelling

1. Fetch images from tabular data (optional as the shared dataset already has fetched images from sdss database) : Image modelling\Fetch images.ipynb
2. SVM : .\Image modelling\SVM.ipynb
3. CNN : .\Image modelling\CNN.ipynb
4. DNN : .\Image modelling\DNN.ipynb
5. ViT : .\Image modelling\ViT Train and Eval.ipynb

Imbalanced techniques

1. Generate quasar augmentations : .\Imbalanced techniques\Augmentation\image_augmentation.ipynb
2. CNN on augmented data : .\Imbalanced techniques\Augmentation\CNN_Augmentations.ipynb
3. Transfer learning on all data : .\Imbalanced techniques\Transfer Learning\All data\all_data_DNN.ipynb
4. Transfer learning on balanced data : .\Imbalanced techniques\Transfer Learning\Balanced data\post_DNN.ipynb
5. SimCLR Pretraining : .\Imbalanced techniques\Pretraining\SimCLR Train.ipynb
6. SimCLR + other classifiers train,eval : .\Imbalanced techniques\Pretraining\SimCLR Validation.ipynb
7. BYOL Pretraining : .\Imbalanced techniques\Pretraining\BYOL Train.ipynb
8. BYOL + other classifiers train,eval : .\Imbalanced techniques\Pretraining\BYOL Validation.ipynb

Alternate (and easily) reproduce results

We also provide Colab/Kaggle versions of most of above notebooks so that reproduction of results gets easier (below are the links).

Image Modelling

1. Fetch images from tabular data (optional as the shared dataset already has fetched images from sdss database) : https://www.kaggle.com/code/rainmaker29/sloan-dataset-code-amaan/notebook
2. CNN : https://colab.research.google.com/drive/1rSXBMaxN1MXrGxNMDwEnVcipveR0iUa3?usp=sharing
3. ViT : https://www.kaggle.com/rainmaker29/vision-transformer-vit-tutorial-baseline

Imbalanced techniques

1. Generate quasar augmentations : https://colab.research.google.com/drive/19Mhhhs0O73tVKQHSW_anubQ0GXepfRIa?usp=sharing
2. CNN on augmented data : https://colab.research.google.com/drive/1f7uU4YYBOJuAw66hR_r4fbSpV5L2PxtD?usp=sharing
3. SimCLR Pretraining : https://www.kaggle.com/rainmaker29/simclr-ml701
4. SimCLR + other classifiers train,eval : https://www.kaggle.com/code/rainmaker29/simclr-post-training-ml701-byol
5. BYOL Pretraining : https://www.kaggle.com/rainmaker29/contrastive-learning-using-byol
6. BYOL + other classifiers train,eval : https://www.kaggle.com/code/mohammadamaansayeed/contrastive-learning-using-byol