# python code/classification/train.py -c classification/config/EL-ResNet18.txt

# Output settings
exp_name = EL-ResNet18
#output_dir = classification/output

# General settings
device = cuda:0
dtype = float32
seed = 1
#load_checkpoint = classification/output/EL-ResNet18.pth

# General training hyperparameters
num_epochs = 200
batch_size = 128
lr = 1e-1
weight_decay = 5e-4
optimizer = RiemannianSGD
use_lr_scheduler = True
lr_scheduler_milestones = [60,120,160]
lr_scheduler_gamma = 0.2

# General validation/testing hyperparameters
batch_size_test = 128

# Model selection
num_layers = 18
embedding_dim = 512
encoder_manifold = euclidean
decoder_manifold = lorentz

# Manifold settings
# learn_k = True
encoder_k = 1.0
decoder_k = 1.0
clip_features = 1.0 # Tiny-ImageNet: clip_features=4.0

# Dataset settings
dataset = CIFAR-100 # CIFAR-10 or Tiny-ImageNet
