Stellar label estimation using deep learning.

This repository contains code to estimate stellar labels.

Eksperiments

In order to run the code, change name of datafile in dataloader.py

Unsupervised learning models

cd src
python denoising.py --name DAE --latentdim 10 --learningrate 3e-4 --num_epochs 750 --model_type DAE1d --batch_size 256 --noise_level 0.01

Supervised learning models

Simple resnet

cd src
python main.py --name MODELNAME --learningrate 1e-4 --convolutions [128,128,128,128,128] --kernel_size [7,7,7,7,7] --epochs 500 --hiddenlayer [1024,1024,512] --model_type bayes --dropout 0.3 --SGD TRUE --noise_level 0.01

Model with denoising

cd src
python3 main.py --name noisefreeresenet --learningrate 1e-4 --convolutions [128,128,128,128,128] --kernel_size [7,7,7,7,7] --epochs 500 --hiddenlayer [1024,1024,512] --model_type bayes --dropout 0.3 --SGD TRUE --noise_level 0.01 --denoise_latent 10 --denoise DAE1d

Attention model

cd src
python3 main.py --name attentionnetwork --learningrate 1e-4--epochs 500 --hiddenlayer [1024,1024,512] --model_type bAttnVGG --batch_size 128 --noise_level 0.01 --SGD TRUE