为解决临床医学量表数据类别不均衡容易对模型产生影响,以及在处理量表数据任务时深度学习框架性能难以媲美传统机器学习方法问题,提出了一种基于级联欠采样的Transformer网络模型(layer by layer Transformer,LLT)。LLT通过级联欠采样...为解决临床医学量表数据类别不均衡容易对模型产生影响,以及在处理量表数据任务时深度学习框架性能难以媲美传统机器学习方法问题,提出了一种基于级联欠采样的Transformer网络模型(layer by layer Transformer,LLT)。LLT通过级联欠采样方法对多数类数据逐层删减,实现数据类别平衡,降低数据类别不均衡对分类器的影响,并利用注意力机制对输入数据的特征进行相关性评估实现特征选择,细化特征提取能力,改善模型性能。采用类风湿关节炎(RA)数据作为测试样本,实验证明,在不改变样本分布的情况下,提出的级联欠采样方法对少数类别的识别率增加了6.1%,与常用的NEARMISS和ADASYN相比,分别高出1.4%和10.4%;LLT在RA量表数据的准确率和F 1-score指标上达到了72.6%和71.5%,AUC值为0.89,mAP值为0.79,性能超过目前RF、XGBoost和GBDT等主流量表数据分类模型。最后对模型过程进行可视化,分析了影响RA的特征,对RA临床诊断具有较好的指导意义。展开更多
The Gaoshan gold-silver deposit, located between the Yuyao-Lishui Fault and Jiangshan- Shaoxing fault in Longquan Area, occurs in the Suichang-Longquan gold-silver polymetallic metallogenic belt. This study conducted ...The Gaoshan gold-silver deposit, located between the Yuyao-Lishui Fault and Jiangshan- Shaoxing fault in Longquan Area, occurs in the Suichang-Longquan gold-silver polymetallic metallogenic belt. This study conducted an investigation for ore-forming fluids using microthermometry, D-O isotope and trace element. The results show that two types of fluid inclusions involved into the formation of the deposit are pure liquid phase and gas-liquid phase aqueous inclusions. The homogenization temperature and salinity of major mineralization phase ranges from 156~C to 236~C （average 200~C） and 0.35% to 8.68% （NaCleqv） （average 3.68%）, respectively, indicating that the ore-forming fluid is characteristic of low temperature and low salinity. The ore- forming pressure ranges between in 118.02 to 232.13＂105 pa, and it is estabmiated that the ore- forming depth ranges from 0.39 to 0.77 km, indicating it is a hypabyssal deposit in genesis. The low rare earth elements content in pyrites, widely developed fluorite in late ore-forming stage and lack of chlorargyrite （AgCI）, indicates that the ore-forming fluid is rich in F rather than CI. The ratios of Y/ Ho, Zr/Hf and Nb/Ta of between different samples have little difference, indicating that the later hydrothermal activities had no effects on the former hydrothermal fluid. The chondrite-normalized REE patterns of pyrites from country rocks and ore veins are basically identical, with the characteristics of light REE enrichment and negative Eu anomalies, implying that the ore-forming fluid was oxidative and derived partly from the country rocks. The JD and jlSo of fluid inclusions in quartz formed during the main metallogenic stage range from -105%o to -69 %0 and -6.01%o to -3.81%o, respectively. The D-O isotopic diagram shows that the metallogenic fluid is characterized by the mixing of formation water and meteoric water, without involvement of magmatic water. The geological and geochemical characteristics of the Gaoshan gold-silver deposit are similar to those of continental volcanic hydrothermal deposit, and could be assigned to the continental volcanic hydrothermal gold-silver deposit type.展开更多
文摘为解决临床医学量表数据类别不均衡容易对模型产生影响,以及在处理量表数据任务时深度学习框架性能难以媲美传统机器学习方法问题,提出了一种基于级联欠采样的Transformer网络模型(layer by layer Transformer,LLT)。LLT通过级联欠采样方法对多数类数据逐层删减,实现数据类别平衡,降低数据类别不均衡对分类器的影响,并利用注意力机制对输入数据的特征进行相关性评估实现特征选择,细化特征提取能力,改善模型性能。采用类风湿关节炎(RA)数据作为测试样本,实验证明,在不改变样本分布的情况下,提出的级联欠采样方法对少数类别的识别率增加了6.1%,与常用的NEARMISS和ADASYN相比,分别高出1.4%和10.4%;LLT在RA量表数据的准确率和F 1-score指标上达到了72.6%和71.5%,AUC值为0.89,mAP值为0.79,性能超过目前RF、XGBoost和GBDT等主流量表数据分类模型。最后对模型过程进行可视化,分析了影响RA的特征,对RA临床诊断具有较好的指导意义。
基金funded by “Preliminary Study On the Metallogenic Conditions and Prospecting Direction of Gold-Silver Deposits,Suichang-Longquan Area,Zhejiang(No.:YK1401)”“Summary and Research Project of the Mineral Geology of China by Mineral Type(Group)(No.:12120114039601)”+1 种基金“Research Project of the Metallogenic Regularity of the National Important Mineral Areas(No.:1212011121037)”“Comprehensive Research Project of China’s Mineral Geology and Regional Metallogenic Regularity(China’s Mineral Geology)(No.:1212011220369)”
文摘The Gaoshan gold-silver deposit, located between the Yuyao-Lishui Fault and Jiangshan- Shaoxing fault in Longquan Area, occurs in the Suichang-Longquan gold-silver polymetallic metallogenic belt. This study conducted an investigation for ore-forming fluids using microthermometry, D-O isotope and trace element. The results show that two types of fluid inclusions involved into the formation of the deposit are pure liquid phase and gas-liquid phase aqueous inclusions. The homogenization temperature and salinity of major mineralization phase ranges from 156~C to 236~C （average 200~C） and 0.35% to 8.68% （NaCleqv） （average 3.68%）, respectively, indicating that the ore-forming fluid is characteristic of low temperature and low salinity. The ore- forming pressure ranges between in 118.02 to 232.13＂105 pa, and it is estabmiated that the ore- forming depth ranges from 0.39 to 0.77 km, indicating it is a hypabyssal deposit in genesis. The low rare earth elements content in pyrites, widely developed fluorite in late ore-forming stage and lack of chlorargyrite （AgCI）, indicates that the ore-forming fluid is rich in F rather than CI. The ratios of Y/ Ho, Zr/Hf and Nb/Ta of between different samples have little difference, indicating that the later hydrothermal activities had no effects on the former hydrothermal fluid. The chondrite-normalized REE patterns of pyrites from country rocks and ore veins are basically identical, with the characteristics of light REE enrichment and negative Eu anomalies, implying that the ore-forming fluid was oxidative and derived partly from the country rocks. The JD and jlSo of fluid inclusions in quartz formed during the main metallogenic stage range from -105%o to -69 %0 and -6.01%o to -3.81%o, respectively. The D-O isotopic diagram shows that the metallogenic fluid is characterized by the mixing of formation water and meteoric water, without involvement of magmatic water. The geological and geochemical characteristics of the Gaoshan gold-silver deposit are similar to those of continental volcanic hydrothermal deposit, and could be assigned to the continental volcanic hydrothermal gold-silver deposit type.
