| File name | Description | Comment |
|---|---|---|
| build_model.py | Model builder by Lenet5 and Net | ipynb is jupyter version (Net version only) |
| build_model_ica.py | Model builder by ICA | Not completed |
| build_model_lenet.ipynb | Jupyter version Lenet5 Model builder | |
| dot_age_full.txt | Full age <-> group mapping | Original dot_age.txt |
| dot_age_test.txt | Testing set of age <-> group mapping | used by predict.py: generate |
| dot_group.txt | Training set of subject mapping | dot_subject.txt + dot_group.txt = dot_age_full.txt |
| dot_subjects_full.txt | Full age <-> group mapping | |
| dot_subjects_test.txt | Test set of age <-> group mapping | |
| dot_subjects.txt | Training set of subject mapping | dot_subject.txt + dot_group.txt = dot_age_full.txt |
| lenet5.pt | Saved lenet5 model | |
| net.pt | Saved net model | |
| predict.py | Perform predicting by unknown testing set | |
| randomize.py | Separate original date to 2 set:Training set / Testing set |
For archive the age group mapping target, the fMRI data are prepossessed, and the fMRI data and the functional connectivity (FNC) matrix between each brain area are calculated respectively.
We use UMCP ( https://github.com/jbrown81/umcp ) tool for calculating this data by transferring our nii data of fMRI 4D
The output is a npy file contains 165 x 165 symmetric matrix.
Here is example of 3x3:
0.000000000000000000e+00 8.684753352306536778e-01 8.475200405138061388e-01
8.684753352306537888e-01 0.000000000000000000e+00 9.139195847122992822e-01
8.475200405138061388e-01 9.139195847122992822e-01 0.000000000000000000e+00
The diagonal element is zero in this matrix. You cannot read npy file directly because it’s for more precise processing by Numpy, so it is not plain text.
For randomize our input, we use randomize.py for split training set and testing set, ratio is 80/20 (TEST_SAMPLE_RATE variable in randomize.py)
We have 152 samples, it means training set will be 121 and testing set is 31
There order has been snuffled in script:
a. open dot_group_full.txt and dot_subject_full.txt
6 with open('dot_subjects_full.txt', 'r') as subject, open('dot_group_full.txt', 'r') as group:
b. store all subject <-> group mapping
7 for subline, groupline in zip(subject, group):
8 train_index[subline.strip('\\n')] = groupline.strip('\\n')