wayfear / brainnetworktransformer Goto Github PK

为什么训练精度为56，测试精度为70+呢？
是我的程序跑错了还是什么问题呢？
希望得到您的回复，感激不尽！

Hello, I'd like to ask two questions regarding the 'abide.npy' dataset.

1. Is 'abide.npy' obtained after processing data following the command git clone https://github.com/preprocessed-connectomes-project/abide and using python download_abide_preproc.py -d rois_cc200 -p cpac -s filt_global -o data/?

2. How are the dimensions of 'final_timeseires' in ABIDE (1009, 200, 100) and 'final_pearson' (1009, 200, 200) determined, and what do they represent?

Thank you.

您的代码中对数据集进行了训练测试验证的划分，想请教您通常这个过程是训练集上训练，调整参数使得验证集上表现最好，然后进行测试集的测试？
还是训练集上训练，测试集上测试，拿到测试集上表现最好的模型，参数保存下来，用这个模型去测试集上测试？

希望能得到您的解答，感激不尽

Hi there,
I'm wondering if this code is compatible with running on multiple GPUs on a single machine.
Could you please clarify?

Thanks you.

Dear authors:
Recently I have looking at your this project and feel interested in it! And the URL is: https://github.com/Wayfear/BrainNetworkTransformer. I have faced some questions, and ask for your help. The first one is that, the installed package pytorch is the gpu version? And another question, is that, this command in the github:

"python -m source --multirun datasz=100p model=bnt,fbnetgen,brainnetcnn,transformer dataset=ABIDE,ABCD repeat_time=5 preprocess=mixup"

I do not know where the source file and the why could I give these paramters, "multirun datasz, model, dataset ", because now I
have not find the source file, and feel confused why could we give these input paramters.
Thank you. I am looking forward to your reply Sincerely!

Best regards.

Could you tell me what "re-standardized" you have done to FC matrix?
I tried calculate the FC matrix using the given time series, but it dosen't match.
I also tried z-score, fisherz and other normalization methods, and it dosen't match too.

The following is how I calculate the FC matrix.
final_pearson[0]

array([[0.        , 0.10491255, 0.33108106, ..., 0.30959536, 0.25176847,
        0.33942461],
       [0.10491255, 0.        , 0.30768242, ..., 0.45559988, 0.18053977,
        0.44306156],
       [0.33108106, 0.30768242, 0.        , ..., 0.30780082, 0.44480115,
        0.14797475],
       ...,
       [0.30959536, 0.45559988, 0.30780082, ..., 0.        , 0.49294844,
        0.54227517],
       [0.25176847, 0.18053977, 0.44480115, ..., 0.49294844, 0.        ,
        0.23103813],
       [0.33942461, 0.44306156, 0.14797475, ..., 0.54227517, 0.23103813,
        0.        ]])

np.corrcoef(final_timeseires[0])

array([[1.        , 0.03009237, 0.29787781, ..., 0.24784874, 0.24297036,
        0.27227381],
       [0.03009237, 1.        , 0.30083494, ..., 0.38774466, 0.07393758,
        0.44119778],
       [0.29787781, 0.30083494, 1.        , ..., 0.10456398, 0.31828787,
        0.10736196],
       ...,
       [0.24784874, 0.38774466, 0.10456398, ..., 1.        , 0.38242007,
        0.50195613],
       [0.24297036, 0.07393758, 0.31828787, ..., 0.38242007, 1.        ,
        0.21774203],
       [0.27227381, 0.44119778, 0.10736196, ..., 0.50195613, 0.21774203,
        1.        ]])

您好，我想问一下文章中说使用了成对注意力，但代码中似乎是计算了自注意力，并且文章的图中表示使用的是缩点注意力，这个是怎么回事呢？

Hi there,

I must say, you did an impressive research work. I was wondering if you could share the .npy and .txt data for ABCD (which you mentioned in ABCD.yml file) and the NIH toolbox data as well. This would be a great help. I am trying to run your code and I have a very short time to do so. Your prepared data would save a lot of time for me.

Thanks a lot. Eagerly waiting to hear from you.

Sorry to bother you, but I have some questions about the heatmaps in the paper and would like to ask. If you don't mind, I hope you can tell me what data Figure3(b) is drawn with.