Table 6 Results of training the subset 1 with different train and test dataset sizes
From: DECIMER 1.0: deep learning for chemical image recognition using transformers
No. | Train data size | Test data size | Split | Average time per epoch | Average Tanimoto | Tanimoto 1.0 (%) |
|---|---|---|---|---|---|---|
1 | 102,400 | 921,600 | 10|90 | 42.22 | 0.86 | 45.05 |
2 | 204,800 | 819,200 | 20|80 | 69.95 | 0.91 | 63.59 |
3 | 307,200 | 716,800 | 30|70 | 199.52 | 0.93 | 71.63 |
4 | 409,600 | 614,400 | 40|60 | 276.09 | 0.94 | 73.93 |
5 | 512,000 | 512,000 | 50|50 | 320.25 | 0.95 | 77.37 |
6 | 614,400 | 409,600 | 60|40 | 392.51 | 0.96 | 84.50 |
7 | 716,800 | 307,200 | 70|30 | 448.91 | 0.97 | 85.38 |
8 | 819,200 | 204,800 | 80|20 | 535.57 | 0.96 | 82.89 |
9 | 921,600 | 102,400 | 90|10 | 560.47 | 0.94 | 75.06 |