trained quartet models for inference of phylogenetically informative features

models_maintrained on rRNA or simulated sequences trained with quartet loss, siamese loss or q+s loss30 output nodes, correct reference treemodels_variableOutputNodes: only trained on simulated sequencestrained with quartet loss, siamese loss or q+s losswith 8, 16, 32, or 64 output featurestrained on correct reference tree or on randomized reference tree<b>All data contain for each epoch:</b>weightsreconstructed trees as nexus and imagefor simulated data: motifs from in-silico mutagenesis as npy filequartet scores and lossesfeature vectors<br>

基于核糖体RNA（ribosomal RNA, rRNA）或模拟序列训练的模型，采用四分体损失（quartet loss）、孪生损失（siamese loss）或四分体+孪生混合损失（quartet+siamese loss, q+s loss）进行训练，配置30个输出节点，以正确参考树为训练基准。
可变输出节点模型：仅基于模拟序列训练，采用四分体损失、孪生损失或四分体+孪生混合损失，输出特征维度可选8、16、32或64，训练基准可选择正确参考树或随机化参考树。
<b>所有数据集在每一轮训练（epoch）中均包含以下内容：</b>
模型权重；
以Nexus格式与图像格式存储的重建树；
若为模拟数据集，则包含计算机诱变（in-silico mutagenesis）产生的基序，存储为.npy文件；
四分体得分与损失值；
特征向量。