mahmoodlab
CONCH encoder embedding
I want to use the CONCH encoder, and I used the following script:
CUDA_VISIBLE_DEVICES=0 python main.py --drop_out 0.25 --early_stopping --lr 2e-4 --k 10 --exp_code task_1_tumor_vs_normal_CLAM_conch_sb --weighted_sample --bag_loss ce --inst_loss svm --task task_1_tumor_vs_normal --model_type clam_sb --log_data --data_root_dir features_conch/ --embed_dim 512
Why is the following error occurring?
Load Dataset
label column: label
label dictionary: {'normal': 0, 'tumor': 1}
number of classes: 2
slide-level counts:
label
1 44
0 34
Name: count, dtype: int64
Patient-LVL; Number of samples registered in class 0: 34
Slide-LVL; Number of samples registered in class 0: 34
Patient-LVL; Number of samples registered in class 1: 44
Slide-LVL; Number of samples registered in class 1: 44
split_dir: splits/task_1_tumor_vs_normal_100
################# Settings ###################
num_splits: 10
k_start: -1
k_end: -1
task: task_1_tumor_vs_normal
max_epochs: 200
results_dir: ./results
lr: 0.0002
experiment: task_1_tumor_vs_normal_CLAM_conch_sb
reg: 1e-05
label_frac: 1.0
bag_loss: ce
seed: 1
model_type: clam_sb
model_size: small
use_drop_out: 0.25
weighted_sample: True
opt: adam
bag_weight: 0.7
inst_loss: svm
B: 8
split_dir: splits/task_1_tumor_vs_normal_100
Training Fold 0!
Init train/val/test splits... Done! Training on 64 samples Validating on 7 samples Testing on 7 samples
Init loss function... Done!
Init Model... Setting tau to 1.0 Done! CLAM_SB( (attention_net): Sequential( (0): Linear(in_features=512, out_features=512, bias=True) (1): ReLU() (2): Dropout(p=0.25, inplace=False) (3): Attn_Net_Gated( (attention_a): Sequential( (0): Linear(in_features=512, out_features=256, bias=True) (1): Tanh() (2): Dropout(p=0.25, inplace=False) ) (attention_b): Sequential( (0): Linear(in_features=512, out_features=256, bias=True) (1): Sigmoid() (2): Dropout(p=0.25, inplace=False) ) (attention_c): Linear(in_features=256, out_features=1, bias=True) ) ) (classifiers): Linear(in_features=512, out_features=2, bias=True) (instance_classifiers): ModuleList( (0-1): 2 x Linear(in_features=512, out_features=2, bias=True) ) (instance_loss_fn): SmoothTop1SVM() ) Total number of parameters: 528647 Total number of trainable parameters: 528647
Init optimizer ... Done!
Init Loaders... Done!
Setup EarlyStopping... Done!
Traceback (most recent call last):
File "/data2/project/CLAM-master/main.py", line 213, in
No, you have used either Resnet50 or UNI to encode, as CONCH is the only one using a 512 dimensional embedding.
