YAD_STAGIN

How to use

please refer the utils/option.py file

python main.py --dataset=hcp_rest --window_siz=50 --window_stride=3 --readout=sero --target=Gender

Hyperparameter optimization

• window length: [25, 50]
• window stride: [1, 2, 3]
• hidden dimension
• sparsity: [10, 20, 30, 40, 50]
• lr: [5e-4]
• max_lr: [1e-3]
• reg_lambda: [1e-5]
• num layer: [2, 3, 4, 5]
• num_head: [1, 3, 5]
• num_epochs: [30, 50]
• minibatch size: [3]
• readout: ['garo', 'sero', 'mean']
• cls_token: ['sum', 'mean', 'param']

Tasks

• classification: Gender, MaDE
• regression: PHQ-9

Current performance

Gender - Scaefer100_sub19	accuracy	recall	precision	roc_auc
yad_rest_schaefer100_sub19_Gender_garo_win30_stride1	0.928±0.027	0.890±0.037	0.928±0.039	0.962±0.020
yad_rest_schaefer100_sub19_Gender_garo_win30_stride1_exceptsSamsung	0.942±0.018	0.905±0.031	0.957±0.035	0.973±0.029
yad_rest_schaefer100_sub19_Gender_sero_win30_stride1	0.884±0.074	0.890±0.058	0.846±0.117	0.936±0.053
yad_rest_schaefer100_sub19_Gender_sero_win30_stride1_exceptsSamsung	0.890±0.029	0.863±0.090	0.880±0.052	0.957±0.030

MaDE - Scaefer100_sub19	accuracy	recall	precision	roc_auc
yad_rest_schaefer100_sub19_MaDE_garo_win30_stride1	0.920±0.013	0.861±0.060	0.909±0.052	0.960±0.006
yad_rest_schaefer100_sub19_MaDE_garo_win30_stride1_exceptsSamsung	0.931±0.039	0.831±0.067	0.938±0.081	0.961±0.032
yad_rest_schaefer100_sub19_MaDE_sero_win30_stride1	0.912±0.041	0.908±0.044	0.852±0.064	0.962±0.016
yad_rest_schaefer100_sub19_MaDE_sero_win30_stride1_exceptsSamsung	0.936±0.022	0.905±0.086	0.899±0.068	0.959±0.027

Gender - Scaefer400_sub19	accuracy	recall	precision	roc_auc
yad_rest_schaefer400_sub19_Gender_garo_win25_stride1	0.928±0.037	0.910±0.037	0.920±0.083	0.969±0.023
yad_rest_schaefer400_sub19_Gender_garo_win25_stride1_except_rois	0.896±0.026	0.870±0.024	0.877±0.071	0.951±0.023
yad_rest_schaefer400_sub19_Gender_garo_win25_stride2	0.924±0.027	0.900±0.055	0.914±0.060	0.977±0.015
yad_rest_schaefer400_sub19_Gender_garo_win25_stride3	0.880±0.044	0.770±0.068	0.917±0.057	0.952±0.027
yad_rest_schaefer400_sub19_Gender_garo_win50_stride3	0.819±0.037	0.630±0.081	0.892±0.072	0.910±0.050
yad_rest_schaefer400_sub19_Gender_sero_win25_stride1_except_rois	0.928±0.035	0.910±0.058	0.920±0.083	0.966±0.032
yad_rest_schaefer400_sub19_Gender_sero_win25_stride2	0.872±0.032	0.830±0.068	0.847±0.032	0.952±0.018
yad_rest_schaefer400_sub19_Gender_sero_win25_stride3	0.900±0.036	0.780±0.075	0.964±0.046	0.948±0.026
yad_rest_schaefer400_sub19_Gender_sero_win50_stride3	0.843±0.077	0.720±0.175	0.877±0.104	0.929±0.049

MaDE - Scaefer400_sub19	accuracy	recall	precision	roc_auc
yad_rest_schaefer400_sub19_MaDE_garo_win25_stride1	0.916±0.048	0.852±0.095	0.921±0.101	0.933±0.034
yad_rest_schaefer400_sub19_MaDE_garo_win25_stride1_exceptsSamsung	0.919±0.029	0.869±0.042	0.877±0.083	0.944±0.036
yad_rest_schaefer400_sub19_MaDE_garo_win25_stride2	0.920±0.022	0.920±0.060	0.864±0.045	0.962±0.022
yad_rest_schaefer400_sub19_MaDE_garo_win25_stride2_exceptsSamsung	0.954±0.014	0.905±0.058	0.945±0.045	0.980±0.014
yad_rest_schaefer400_sub19_MaDE_garo_win25_stride3	0.928±0.020	0.851±0.044	0.941±0.051	0.965±0.027
yad_rest_schaefer400_sub19_MaDE_sero_win25_stride1	0.916±0.041	0.850±0.060	0.911±0.090	0.946±0.036
yad_rest_schaefer400_sub19_MaDE_sero_win25_stride1_except_rois	0.928±0.054	0.875±0.082	0.919±0.084	0.966±0.037
yad_rest_schaefer400_sub19_MaDE_sero_win25_stride1_exceptsSamsung	0.937±0.028	0.851±0.040	0.944±0.074	0.965±0.036
yad_rest_schaefer400_sub19_MaDE_sero_win25_stride2	0.879±0.044	0.806±0.104	0.850±0.081	0.947±0.028
yad_rest_schaefer400_sub19_MaDE_sero_win25_stride2_exceptsSamsung	0.907±0.038	0.791±0.124	0.891±0.016	0.976±0.017
yad_rest_schaefer400_sub19_MaDE_sero_win25_stride3	0.900±0.013	0.804±0.032	0.903±0.059	0.960±0.019