kvzhao/marshall_rule

Use simple network to learn Marshall sign rule (-1)^N_Asite and beyond

Marshall rule

intro

We study, in this project, how neural networks capturing known phyiscal rules.

usage

Run training (by default)

python main.py 

There are following options

• is_train
• cell_size
• num_layers
• task_name
• DATA_PATH
• LABEL_PATH
• NUM_EPOCH
• BATCH_SIZE

for example

python main.py --is_train=True --cell_size=16 --num_layers=2 --DATA_PATH=datasetMerged/states_J0.txt --LABEL_PATH=datasetMerged/sign_J0.txt --task_name=demo --SAVE_CKPT_PER_STEPS=100000

note: emprically, datasetConfig takes 200k iterations to converge, and J1J2 takes about 100k.

After

python main.py --is_train=False

todo

• ipython notebook for testing visualization
• trained model