Figure 2 - data

Figure2. Training design and performance of ML models. b)LFP example of the new test set and the corresponding model outputs per window of analysis. Note different duration of true events. Setting a threshold allows defining the windows containing detected events. Colored ticks represent detections by the different models. Two different thresholds (dark and light gray) can influence what events are detected. Note how detections marked with arrows are dismissed when the threshold increases. Since SWRs constitute about 1-4% of the total recording duration, performance is computed using positive detections; that is windows without GT or detected events are not computed for performance.

图2 机器学习模型的训练设计与性能表现。(b) 新测试集的局部场电位（Local Field Potential, LFP）示例，以及各分析窗口对应的模型输出。请注意真实事件的持续时长各不相同。通过设定阈值可界定包含检测事件的窗口。彩色刻度代表不同模型的检测结果。两种不同阈值（深灰色与浅灰色）会改变可被检测到的事件范围。请注意当阈值升高时，箭头标记的检测结果会被剔除。由于尖波涟漪（Sharp Wave-Ripples, SWRs）仅占总记录时长的1%~4%，因此模型性能仅基于阳性检测结果计算：即不纳入无真实标签（Ground Truth, GT）或未检测到事件的窗口。