journal menu
![[Figure 6]](qi5003fig6mag.jpg)
|
|
Figure 6
Training optimization curves. (a) Accuracy versus training epoch for the resolution-prediction model. This is a regression model, for which we define accuracy as the fraction of images whose predictions are within 0.07 Å−1 of the ground truth. The training job was stopped after epoch 354 and then restarted, as indicated by the discontinuity. (b) Accuracy versus epoch for the overlapping lattice-detection model. Here, accuracy is the fraction of predictions with the correct label. For both plots, the test curves (black markers) are derived from images that were never used for training. Eventually, training accuracy diverges, indicating model bias. The vertical lines mark the epoch where models were saved for use with experimental data. |
![[Figure 6]](qi5003fig6.jpg)
 | STRUCTURAL BIOLOGY |
ISSN: 2059-7983
access
