initial commit Python 3.7 version
MPNN/ -- model and layers of Message Passing Neural Network
main.sh --running shell for automl
run.sh --running shell for python command line
para.yml --parameters configuration for automl
train.py --program entrance, training and validation
*Model Parameters
--Data argumentation
num_rotate (int): # of coordinates rotation in [0, 2*pi]
num_grid (int): (grid * grid) image for put coordinates
scale_factor (int): reduce the image resolution by scale_factor
flip (bool) : whether to flip image
the final image size is (num_grid * scale_factor, num_gird * scale_factor)
--Model
model_type (str): 'alexnet', 'resnet18', ...
--Training
epoches (int)
learing_rate (float)
batch_size (int)
weight_decay (float): L2 regularization of Adam
decay_factor (float): learning rate decay factor (exp)
decay_patience (int): # epoches for one lr decay
--Other
num_workers (int): # of workers for Dataset