Local connectome fingerprints of HCP 1062 subjects for prediction

We provide a simple model for training/testing the NEO-FFI Prediction Task using PyTorch. For each question, 10-fold cross-valdation will take roughly 2.5 minutes with batch size 16 to train and test on a GTX 1080 Ti GPU machine. The time for experimenting on 60 quesitons is around 2.5 hours. The dataset is available at here. Please contact Fang-Cheng Yeh or Po-Yao Huang if you have any questions.

You are encouraged to develop more advanced model to improve the regression perfromance on diffusion MRI. You may start by modifying model.py or customize the loss function. Any PR is more than welcome.

Usage

python main.py [-h] [--kfold KFOLD] [--lr LR] [--bs BS] [--epoch EPOCH] [--loss LOSS] [--leaky] [--layers LAYERS] [--model_name MODEL_NAME] [--hidden HIDDEN]

Example

For a 3-layer MLP model with batch size 16 and 2e-6 learning rate and l1 loss, the command is:

python main.py --model_name l1_hh128_relu_b2_lr2e6 --loss l1 --lr 0.000002 --layers 3 --bs 16

Please check go.sh for an example script which trains 3 models sequentially.

Visualization

tensorboard --logdir=run/

Analysis

The performance log file is stored at ./logs/[model_name]. We provide a script to show the reslts:

python log/show.py [model_name]

Performance

Model Name:l1_hh128_relu_b16_lr1e5 Metric: Pearson Corr

ID	Mean	STD	Max	Min
0	0.2906	0.0626	0.4469	0.2282
1	0.1591	0.0846	0.2881	0.0214
2	0.1790	0.0744	0.3188	0.0295
3	0.0750	0.0559	0.1791	0.0000
4	0.1548	0.0831	0.2593	0.0000
5	0.1568	0.0372	0.2049	0.0896
6	0.1383	0.0893	0.3318	0.0117
7	0.1432	0.0783	0.2486	0.0212
8	0.2262	0.0705	0.3226	0.1234
9	0.1358	0.0660	0.2427	0.0000
10	0.2062	0.0742	0.3014	0.0482
11	0.1062	0.0751	0.2440	0.0000
12	0.0677	0.0466	0.1535	0.0000
13	0.1849	0.0948	0.3260	0.0490
14	0.1304	0.0407	0.1932	0.0471
15	0.1456	0.0845	0.3223	0.0066
16	0.1327	0.0580	0.2504	0.0379
17	0.1656	0.0587	0.2828	0.0732
18	0.1229	0.0682	0.2388	0.0203
19	0.1041	0.0845	0.2745	0.0000
20	0.1248	0.0446	0.2139	0.0804
21	0.2079	0.0677	0.3425	0.1255
22	0.1171	0.0753	0.2064	0.0034
23	0.1643	0.0777	0.3261	0.0284
24	0.1483	0.0606	0.3129	0.0966
25	0.1424	0.0906	0.3063	0.0000
26	0.1939	0.0751	0.3668	0.1004
27	0.1513	0.0864	0.3198	0.0308
28	0.2959	0.0658	0.4176	0.2155
29	0.2503	0.0398	0.3185	0.1713
30	0.1294	0.0882	0.2591	0.0000
31	0.1811	0.0865	0.2860	0.0254
32	0.1950	0.0834	0.3651	0.0902
33	0.1136	0.0711	0.2347	0.0025
34	0.0925	0.0460	0.1727	0.0182
35	0.2372	0.0794	0.3468	0.1087
36	0.1140	0.0448	0.2065	0.0250
37	0.2456	0.0918	0.3746	0.1146
38	0.2886	0.0411	0.3478	0.2254
39	0.1038	0.0865	0.2118	0.0000
40	0.1596	0.0384	0.2228	0.0955
41	0.1282	0.0576	0.2475	0.0526
42	0.1474	0.0452	0.2208	0.0788
43	0.1412	0.0717	0.2602	0.0645
44	0.2042	0.0761	0.3545	0.1036
45	0.2363	0.0808	0.4446	0.1395
46	0.1439	0.0768	0.2507	0.0141
47	0.2336	0.0863	0.3848	0.0627
48	0.0841	0.0735	0.2026	0.0000
49	0.1299	0.0664	0.2383	0.0000
50	0.1447	0.0648	0.2452	0.0156
51	0.1558	0.0733	0.2776	0.0325
52	0.2446	0.0992	0.4472	0.1389
53	0.2245	0.0440	0.3206	0.1715
54	0.1446	0.0515	0.2523	0.0689
55	0.1231	0.0546	0.2013	0.0359
56	0.1314	0.0642	0.2210	0.0148
57	0.2150	0.0495	0.2861	0.1302
58	0.1645	0.0561	0.2571	0.0961
59	0.1059	0.0829	0.2619	0.0000