File size: 2,340 Bytes
14dce77 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 |
experiment_name = "unet_global_padding_nov_5_no_lsdb"
base_dir = "/exports/csce/eddie/eng/groups/DunnGroup/matthew/models_gelgenie"
[processing]
base_hardware = "EDDIE"
device = "GPU"
pe = 1
memory = 64
[data]
n_channels = 1
batch_size = 2
num_workers = 1
val_percent = 10
dir_train_mask = [ "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/nathan_gels/masks", "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/matthew_gels/masks", "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/matthew_gels_2/masks", "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/neb_ladders/masks",]
dir_train_img = [ "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/nathan_gels/images", "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/matthew_gels/images", "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/matthew_gels_2/images", "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/neb_ladders/images",]
dir_val_img = [ "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/nathan_gels/val_images", "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/matthew_gels/val_images", "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/matthew_gels_2/val_images", "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/neb_ladders/val_images",]
dir_val_mask = [ "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/nathan_gels/val_masks", "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/matthew_gels/val_masks", "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/matthew_gels_2/val_masks", "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/neb_ladders/val_masks",]
split_training_dataset = false
apply_augmentations = true
padding = true
individual_padding = false
weak_augmentations = false
[model]
model_name = "smp_unet"
classes = 2
in_channels = 1
encoder_name = "resnet18"
[training]
loss = [ "dice", "crossentropy",]
loss_component_weighting = [ 1, 1,]
class_loss_weighting = false
class_loss_weight_damper = [ 1.0, 1.0,]
lr = 0.0001
epochs = 600
grad_scaler = true
load_checkpoint = false
optimizer_type = "adam"
scheduler_type = "CosineAnnealingWarmRestarts"
save_checkpoint = true
checkpoint_frequency = 1
wandb_track = true
model_cleanup_frequency = 20
wandb_id = "2ak0r0kx"
[training.scheduler_specs]
restart_period = 100
|