Custom code created for the purposes of the thesis: "Applications of statistical theory to sensor data analysis"

This is the custom code repository for replicating the results of the thesis. Three main routines are contained within this repository.A new quality measure is proposed in the thesis for the purposes of assessing the quality of predictors in human activity recognition problems. The related code can be found in the file: measures.pyA postprocessing scheme is proposed in the thesis to remove unrealistically short activities from the classification given by the predictor. The related code can be found in the file: postprocessing.pyA new formulation of the null hypothesis in a permutation test for no effect is proposed in the thesis. The viability of the test is presented based on the simulation study. This simulation study can be found in the files: sim_study_lin_reg.ipynb and sim_study_nn.ipynb.

本仓库为复现该论文研究成果的定制化代码仓库，共包含三项核心研究相关代码模块。论文中提出了一种全新的质量度量指标，用于评估人类活动识别任务中预测器的性能质量，相关代码存放于measures.py文件中；此外，论文还提出了一种后处理方案，用于从预测器输出的分类结果中移除不符合实际的短时长活动片段，相关代码存放于postprocessing.py文件中；最后，论文针对无效应置换检验提出了一种全新的原假设（null hypothesis）构建方式，并通过模拟研究验证了该检验的可行性，相关模拟研究代码存放于sim_study_lin_reg.ipynb与sim_study_nn.ipynb文件中。