Student Asks Whether to Publish QSPR Deep Learning Model with 0.64 R² on Melting Point Prediction
English summary
A university student developed a random forest model (r²=0.66, file size 1.23 GB) and a smaller PyTorch deep learning model (270k parameters, 1.3–1.4 MB, r²=0.64) for predicting melting points of chemical compounds using topological indices from the Jean-Claude Bradley Open Melting Point Dataset. The deep learning model achieved MAE 41.25 K, RMSE 54.67 K, and MAPE 11.69%. The student solicits community advice on whether to commit and publish these results or continue trying to improve the model.
Chinese summary
一名大学生基于拓扑指数和Jean-Claude Bradley开放熔点数据集，训练了随机森林模型（R²=0.66，文件大小1.23 GB）和更小的PyTorch深度学习模型（27万参数，1.3–1.4 MB，R²=0.64）来预测化合物熔点。该深度学习模型达到MAE 41.25 K、RMSE 54.67 K、MAPE 11.69%。学生向社区询问是应该以此成果投稿发表，还是继续改进模型。
Key points
Random forest model: r²=0.66, file size 1.23 GB.
随机森林模型：R²=0.66，文件大小1.23 GB。
Custom deep learning model: 270k parameters, 1.3–1.4 MB, r²=0.64, trained with PyTorch.
自定义深度学习模型：27万参数，1.3–1.4 MB，R²=0.64，使用PyTorch训练。
Evaluation metrics on test set: MAE 41.25 K, RMSE 54.67 K, MAPE 11.69%.
测试集评价指标：MAE 41.25 K，RMSE 54.67 K，MAPE 11.69%。
The post is a request for advice on publishing versus further improvement, not a paper or product announcement.
发帖是为获得发表还是继续改进的建议，并非论文或产品发布。