example_config.toml

[classification]
track_statistics = false
track_predictions = false
track_embeddings = false # tracks badge and coreset embeddings if set to true
track_probs = false
model = 'resnet-18'
loss = 'ce'
pretrained = false
epochs = 2
batch_size = 128
test_batch_size = 128

[classification.optimization]
# active learning
lr = 0.001
optimizer = 'adam'
scheduler = 'multiLR'
multiLR_steps = 2
gamma = 0.1


[uspec_configs]
num_seeds = 5

[run_configs]
output_folder_name = 'run'
ld_folder_name = 'testld_output/'
resume = 'none'
cuda = true
gpu_id = 0
train = false
test = false

# seed range for parallel running
start_seed = 0
end_seed = 1

[run_configs.ood]
svhn = false # only classification
mnist = true # only classification
cifar10 = true # only classification
stl10 = true # only classification
cifar10C = false
cifar100 = false
cinic10 = false

[active_learning]
n_start = 128
n_end = 4000
n_query = 1024
strategy = 'random'
init_seed = 0
convergence_acc = 92.0
save_query_idxs = false


[active_learning.stats]
stat_sampling_type = 'SV'
pcal_sampling_type = 'pc'
flip_sampling_type = 'bc'
eventsampling_file = '~/INSERTPATH'
probsampling_file = '~/INSERTPATH'
embedsampling_file = '~/INSERTPATH'
secondary_samping_type = 'entropy'
relaxation = 1024

[data]
dataset = 'CIFAR10' # choice between MNIST, CIFAR10 etc.
data_loc = '~/data/'
download = false # specifies wether to download the dataset or not

[data.augmentations]
random_crop = false
random_hflip = false
random_rotate = false # only segmentation
cutout = false # only classification