Open DragonEmperorG opened 10 months ago
在Google Smartphone Decimeter Challenge 2023数据集上将一条轨迹整个作为一段时间序列数据送入网络进行训练,验证集Loss不下降的问题,这样窗口参数的选择可能是问题的原因。整个网络的参数不足以支持超长时间序列的预测,减小窗口参数的大小,产生更多的数据段使得数据规模和网络参数能够匹配
在Google Smartphone Decimeter Challenge 2023数据集上将一条轨迹整个作为一段时间序列数据送入网络进行训练,验证集Loss不下降的问题,这样窗口参数的选择可能是问题的原因。整个网络的参数不足以支持超长时间序列的预测,减小窗口参数的大小,产生更多的数据段使得数据规模和网络参数能够匹配