Here we decided to set the cutoff high at 50 bits, as depicted in Figure 5, in accordance to the maximum prediction accuracy and to diminish as much as possible the rate of false positives included in the predictions.
Figure 1 a,b) demonstrates the adjusted R distributions for these 4,095 mESC TF models and 255 mESC HM+DNase models respectively, with the minimum and maximum prediction accuracies for each n and m connected by the blue and red curves.
The maximum prediction uncertainties of FIA and Lefebvreʼs model are ±20%and±50%0%, respectively.
The minimum and maximum prediction results are labeled in the form of error bars on the top of the blue bars.
We confirm that soft targets from the output distribution of the teacher are more informative for the student than hard targets based on the class having the maximum prediction of the teacher and provide some interesting insight into the weight matrices learned by soft target training which suggests they tend to be fuller rank as a result of the richer information in soft targets.
For classifiers trained on feature sets ranging from 1 to 1000 features, we plotted the maximum prediction accuracies of transcript- and marker-based prediction for each SMP; more or fewer features are required for maximum accuracy depending on the SMP.
The maximum prediction errors for SR1, SR2 and SR3 are, respectively, 12.1, 28.7 and 12.4%.
The maximum prediction errors for SC1, SC2 and SC3 are, respectively, 11.1, 12.7 and 27.5%.
We can see the predictions clustering along a diagonal, and the maximum prediction of any bin is approximately 3.5 earthquakes.
The developed model was validated with a new set of experimental data, and the maximum prediction error of the model was less than 7%.
