- Machine Learning with Swift
- Alexander Sosnovshchenko
- 132字
- 2021-06-24 18:55:05
Balanced dataset
The application allows you to record samples of different motion types. As you train the model, you may notice one interesting effect: to get accurate predictions, you need not only enough samples, but you also need the proportion of different classes in your dataset to be roughly equal. Think about it: if you have 100 samples of two classes (walk and run), and 99 of them belong to one class (walk), the classifier that delivers 99% accuracy may look like this:
func predict(x: [Double]) -> MotionType { return .walk }
But this is not what we want, obviously.
This observation lead us to the notion of the balanced data set; for most machine learning algorithms, you want the data set in which samples of different classes are represented equally frequently.
推薦閱讀
- Cortex-M3 + μC/OS-II嵌入式系統開發入門與應用
- 極簡Spring Cloud實戰
- 深入淺出SSD:固態存儲核心技術、原理與實戰
- 電腦常見故障現場處理
- Manage Partitions with GParted How-to
- AMD FPGA設計優化寶典:面向Vivado/SystemVerilog
- 電腦維護365問
- SiFive 經典RISC-V FE310微控制器原理與實踐
- Blender Quick Start Guide
- Spring Cloud微服務和分布式系統實踐
- Wireframing Essentials
- 電腦橫機使用與維修
- IP網絡視頻傳輸:技術、標準和應用
- 筆記本電腦維修技能實訓
- Blender for Video Production Quick Start Guide