| import numpy as np | |
| # arr_outcomes = np.load('../processed_data/arr_outcomes.npy', allow_pickle=True) | |
| # split randomization over folds | |
| """Use 9:1:1 split""" | |
| p_train = 0.80 | |
| p_val = 0.10 | |
| p_test = 0.10 | |
| n = 11988 # original 12000 patients, remove 12 outliers | |
| n_train = round(n*p_train) | |
| n_val = round(n*p_val) | |
| n_test = n - (n_train+n_val) | |
| print(n_train, n_val, n_test) | |
| Nsplits = 5 | |
| for j in range(Nsplits): | |
| p = np.random.permutation(n) | |
| idx_train = p[:n_train] | |
| idx_val = p[n_train:n_train+n_val] | |
| idx_test = p[n_train+n_val:] | |
| np.save('../splits/phy12_split'+str(j+1)+'.npy', (idx_train, idx_val, idx_test)) | |
| # np.save('../splits/phy12_split_subset'+str(j+1)+'.npy', (idx_train, idx_val, idx_test)) | |
| print('split IDs saved') | |
| # # check first split | |
| # idx_train,idx_val,idx_test = np.load('../splits/phy12_split1.npy', allow_pickle=True) | |
| # print(len(idx_train), len(idx_val), len(idx_test)) |