-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathrun_commands.txt
55 lines (36 loc) · 7.2 KB
/
run_commands.txt
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
51
52
53
54
55
--- Pretraining ---
--------- 60s clips w/ saved preproc ---------
python train_ssl.py --input_dir /storage/datasets_public/eeg_tuh_seizure/Resampled_TUH_SZ_v1.5.2 --raw_data_dir /storage/datasets_public/eeg_tuh_seizure/TUH_SZ_v1.5.2 --save_dir /home/users/arash/W4H/eeg/saved_results/pretrained_eeg_ssl/clip-60s --graph_type combined --max_seq_len 60 --output_seq_len 12 --do_train --num_epochs 350 --task SS\ pre-training --metric_name loss --use_fft --lr_init 5e-4 --num_rnn_layers 3 --rnn_units 64 --max_diffusion_step 2 --preproc_dir /storage/datasets_public/eeg_tuh_seizure/Preprocessed_TUH_SZ_v1.5.2/seizure_detection_clip_60/clipLen60_timeStepSize1 --data_augment --device cuda --model_name neurognn --use_wand
--------- 12s clips w/ saved preproc ---------
python train_ssl.py --input_dir /storage/datasets_public/eeg_tuh_seizure/Resampled_TUH_SZ_v1.5.2 --raw_data_dir /storage/datasets_public/eeg_tuh_seizure/TUH_SZ_v1.5.2 --save_dir /home/users/arash/W4H/eeg/saved_results/pretrained_eeg_ssl/clip-12s --graph_type combined --max_seq_len 12 --output_seq_len 12 --do_train --num_epochs 350 --task SS\ pre-training --metric_name loss --use_fft --lr_init 5e-4 --num_rnn_layers 3 --rnn_units 64 --max_diffusion_step 2 --preproc_dir /storage/datasets_public/eeg_tuh_seizure/Preprocessed_TUH_SZ_v1.5.2/seizure_detection_clip_12/clipLen12_timeStepSize1 --data_augment --device cuda --model_name neurognn --use_wand
--------- 60s clips w.o/ saved preproc ---------
python train_ssl.py --input_dir /storage/datasets_public/eeg_tuh_seizure/Resampled_TUH_SZ_v1.5.2 --raw_data_dir /storage/datasets_public/eeg_tuh_seizure/TUH_SZ_v1.5.2 --save_dir /home/users/arash/W4H/eeg/saved_results/pretrained_eeg_ssl/clip-60s --graph_type combined --max_seq_len 60 --output_seq_len 12 --do_train --num_epochs 350 --task SS\ pre-training --metric_name loss --use_fft --lr_init 5e-4 --num_rnn_layers 3 --rnn_units 64 --max_diffusion_step 2 --data_augment --device cuda --model_name neurognn --use_wand
--------- 12s clips w.o/ saved preproc ---------
python train_ssl.py --input_dir /storage/datasets_public/eeg_tuh_seizure/Resampled_TUH_SZ_v1.5.2 --raw_data_dir /storage/datasets_public/eeg_tuh_seizure/TUH_SZ_v1.5.2 --save_dir /home/users/arash/W4H/eeg/saved_results/pretrained_eeg_ssl/clip-12s --graph_type combined --max_seq_len 12 --output_seq_len 12 --do_train --num_epochs 350 --task SS\ pre-training --metric_name loss --use_fft --lr_init 5e-4 --num_rnn_layers 3 --rnn_units 64 --max_diffusion_step 2 --data_augment --device cuda --model_name neurognn --use_wand
--- Classification and Detection without fine-tuning with saved preproc (remove --preproc_dir if not saved) ---
---- UPDATED COMMAND ----
Classification (60s clips)
```
python train.py --input_dir /storage/datasets_public/eeg_tuh_seizure/Resampled_TUH_SZ_v1.5.2 --raw_data_dir /storage/datasets_public/eeg_tuh_seizure/TUH_SZ_v1.5.2 --save_dir /home/users/arash/W4H/eeg/saved_results/seizure_classification_clip_60 --graph_type combined --max_seq_len 60 --do_train --num_epochs 60 --task classification --metric_name F1 --use_fft --lr_init 2e-4 --num_rnn_layers 2 --max_diffusion_step 2 --num_classes 4 --data_augment --device cuda --preproc_dir /storage/datasets_public/eeg_tuh_seizure/Preprocessed_TUH_SZ_v1.5.2/seizure_classification_clip_60/clipLen60_timeStepSize1 --model_name neurognn --dropout 0.5
```
(Need to replace save_dir with the user's desired save directory)
Detection (60s clips)
```
python train.py --input_dir /storage/datasets_public/eeg_tuh_seizure/Resampled_TUH_SZ_v1.5.2 --raw_data_dir /storage/datasets_public/eeg_tuh_seizure/TUH_SZ_v1.5.2 --save_dir /home/users/arash/W4H/eeg/saved_results/seizure_detection_clip_60 --graph_type combined --max_seq_len 60 --do_train --num_epochs 100 --task detection --metric_name auroc --use_fft --lr_init 1e-4 --num_rnn_layers 2 --max_diffusion_step 2 --num_classes 1 --data_augment --device cuda --preproc_dir /storage/datasets_public/eeg_tuh_seizure/Preprocessed_TUH_SZ_v1.5.2/seizure_detection_clip_60/clipLen60_timeStepSize1 --model_name neurognn --use_wandb
```
To fine-tune the retrained model, add the following arguments to the run commands:
```
--fine_tune --load_model_path /home/users/arash/W4H/eeg/NeuroGNN/pretrained/pretrained_distance_graph_60s.pth.tar --pretrained_num_rnn_layers 3
```
(Need to replace load_model_path with the user's pretrained checkpoint)
--- Legacy Command ---
--------- 60s clip detection ---------
python train.py --input_dir /storage/datasets_public/eeg_tuh_seizure/Resampled_TUH_SZ_v1.5.2 --raw_data_dir /storage/datasets_public/eeg_tuh_seizure/TUH_SZ_v1.5.2 --save_dir /home/users/arash/W4H/eeg/saved_results/seizure_detection_clip_60 --graph_type combined --max_seq_len 60 --do_train --num_epochs 100 --task detection --metric_name auroc --use_fft --lr_init 1e-4 --num_rnn_layers 2 --max_diffusion_step 2 --num_classes 1 --data_augment --device cuda --preproc_dir /storage/datasets_public/eeg_tuh_seizure/Preprocessed_TUH_SZ_v1.5.2/seizure_detection_clip_60/clipLen60_timeStepSize1 --model_name neurognn --use_wandb
--------- 12s clip detection ---------
python train.py --input_dir /storage/datasets_public/eeg_tuh_seizure/Resampled_TUH_SZ_v1.5.2 --raw_data_dir /storage/datasets_public/eeg_tuh_seizure/TUH_SZ_v1.5.2 --save_dir /home/users/arash/W4H/eeg/saved_results/seizure_detection_clip_12 --graph_type combined --max_seq_len 12 --do_train --num_epochs 100 --task detection --metric_name auroc --use_fft --lr_init 1e-4 --num_rnn_layers 2 --max_diffusion_step 2 --num_classes 1 --data_augment --device cuda --preproc_dir /storage/datasets_public/eeg_tuh_seizure/Preprocessed_TUH_SZ_v1.5.2/seizure_detection_clip_12/clipLen12_timeStepSize1 --model_name neurognn --use_wandb
--------- 60s clip classification ---------
python train.py --input_dir /storage/datasets_public/eeg_tuh_seizure/Resampled_TUH_SZ_v1.5.2 --raw_data_dir /storage/datasets_public/eeg_tuh_seizure/TUH_SZ_v1.5.2 --save_dir /home/users/arash/W4H/eeg/saved_results/seizure_classification_clip_60 --graph_type combined --max_seq_len 60 --do_train --num_epochs 60 --task classification --metric_name F1 --use_fft --lr_init 3e-4 --num_rnn_layers 2 --max_diffusion_step 2 --num_classes 4 --data_augment --device cuda --preproc_dir /storage/datasets_public/eeg_tuh_seizure/Preprocessed_TUH_SZ_v1.5.2/seizure_classification_clip_60/clipLen60_timeStepSize1 --model_name neurognn --use_wandb --dropout 0.5
--------- 12s clip classification ---------
python train.py --input_dir /storage/datasets_public/eeg_tuh_seizure/Resampled_TUH_SZ_v1.5.2 --raw_data_dir /storage/datasets_public/eeg_tuh_seizure/TUH_SZ_v1.5.2 --save_dir /home/users/arash/W4H/eeg/saved_results/seizure_classification_clip_12 --graph_type combined --max_seq_len 12 --do_train --num_epochs 12 --task classification --metric_name F1 --use_fft --lr_init 3e-4 --num_rnn_layers 2 --max_diffusion_step 2 --num_classes 4 --data_augment --device cuda --preproc_dir /storage/datasets_public/eeg_tuh_seizure/Preprocessed_TUH_SZ_v1.5.2/seizure_classification_clip_12/clipLen12_timeStepSize1 --model_name neurognn --use_wandb --dropout 0.5
-- For fine-tuning pre-trained models, add the following to the end (for classificaiton and detection)
--fine_tune --load_model_path /home/users/arash/W4H/eeg/saved_results/pretrained_eeg_ssl/<LOCATION> --pretrained_num_rnn_layers 1