加性分位数回归为非线性关系的建模提供一种灵活、鲁棒的方法.拟合加性分位数模型的方法通常使用样条函数逼近分量,但需要先验的选择节点,计算速度较慢,并不适合大规模数据问题.因此文中提出基于融合Lasso的非参数加性分位数回归模型(No...加性分位数回归为非线性关系的建模提供一种灵活、鲁棒的方法.拟合加性分位数模型的方法通常使用样条函数逼近分量,但需要先验的选择节点,计算速度较慢,并不适合大规模数据问题.因此文中提出基于融合Lasso的非参数加性分位数回归模型(Nonparametric Additive Quantile Regression Model Based on Fused Lasso,AQFL),是在融合Lasso罚和l_(2)罚之间折衷的可对加性分位数回归模型进行估计和变量选择的模型.融合Lasso罚使模型能快速计算,并在局部进行自适应,从而实现对所需分位数甚至极端分位数的预测.同时结合l_(2)罚,在高维数据中将对响应影响较小的协变量函数值压缩为零,实现变量的选择.此外,文中给出保证收敛到全局最优的块坐标ADMM算法(Block Coordinate Alternating Direction Method of Multipliers,BC-ADMM),证明AQFL的预测一致性.在合成数据和碎猪肉数据上的实验表明AQFL在预测准确性和鲁棒性等方面较优.展开更多
Vocabulary knowledge is an important component of the writing skill and it has many dimensions, such as size, depth, and productive, in interaction with writing skill. To evaluate this relation and determine which dim...Vocabulary knowledge is an important component of the writing skill and it has many dimensions, such as size, depth, and productive, in interaction with writing skill. To evaluate this relation and determine which dimension is the most effective for second language writing quality, the present study was conducted. Turkish EFL （English as a Foreign Language） learners＇ lexical competence and writing abilities were examined through their vocabulary profiles and academic essays. The results of each vocabulary measure indicated that the participants had a limited vocabulary size, containing words mostly from 2,000 to 3,000 frequency bands and thus, the productive vocabulary knowledge of the participants mostly consisted of lk ＋ 2k words and the use of academic words in their essays was very low. The results of the study revealed that the lexical competence covering the main components of vocabulary knowledge was a good predictor of the students＇ quality of writing performance.展开更多
文摘加性分位数回归为非线性关系的建模提供一种灵活、鲁棒的方法.拟合加性分位数模型的方法通常使用样条函数逼近分量,但需要先验的选择节点,计算速度较慢,并不适合大规模数据问题.因此文中提出基于融合Lasso的非参数加性分位数回归模型(Nonparametric Additive Quantile Regression Model Based on Fused Lasso,AQFL),是在融合Lasso罚和l_(2)罚之间折衷的可对加性分位数回归模型进行估计和变量选择的模型.融合Lasso罚使模型能快速计算,并在局部进行自适应,从而实现对所需分位数甚至极端分位数的预测.同时结合l_(2)罚,在高维数据中将对响应影响较小的协变量函数值压缩为零,实现变量的选择.此外,文中给出保证收敛到全局最优的块坐标ADMM算法(Block Coordinate Alternating Direction Method of Multipliers,BC-ADMM),证明AQFL的预测一致性.在合成数据和碎猪肉数据上的实验表明AQFL在预测准确性和鲁棒性等方面较优.
文摘Vocabulary knowledge is an important component of the writing skill and it has many dimensions, such as size, depth, and productive, in interaction with writing skill. To evaluate this relation and determine which dimension is the most effective for second language writing quality, the present study was conducted. Turkish EFL （English as a Foreign Language） learners＇ lexical competence and writing abilities were examined through their vocabulary profiles and academic essays. The results of each vocabulary measure indicated that the participants had a limited vocabulary size, containing words mostly from 2,000 to 3,000 frequency bands and thus, the productive vocabulary knowledge of the participants mostly consisted of lk ＋ 2k words and the use of academic words in their essays was very low. The results of the study revealed that the lexical competence covering the main components of vocabulary knowledge was a good predictor of the students＇ quality of writing performance.
