wklix
wklix
我在使用Unimol tools,我将split定义为scaffold,但是我发现结果和random数据划分方式结果一样,请问是我这种直接更改方式不正确吗? clf = MolTrain(task='regression', data_type='molecule', epochs=200, learning_rate=0.00005, batch_size=4, early_stopping=5, split='scaffold', save_path='./scaffold‘, remove_hs=True, )
无法预测出负值
1、请问Unimol tools是完全基于Unimol吗?我在用Unimol tools复现您Unimol论文中回归任务,具体我在esol,freesolv,lipo数据集的RMSE结果是0.776,2.052,0.607,您的论文esol,freesolv,lipo分别是0.788 1.480 0.603。对于freesolv数据集无论怎么调参数都很难达到论文中的指标。 2、在利用Unimol tools进行我个人数据集的回归任务,但我发现无法预测出负值。请问怎么办
Unimol是否可以完成输入是smiles,输出也是smiles,用来预测出新分子
I have noticed a significant disparity between the predicted values and the labels when using Unimol tools for predicting molecular properties, especially when the molecular property values are either large...
请问前辈可以提供一下训练好的.pth的模型文件吗?我的邮箱是[email protected]
Would you please provide the training code and eval code