infinitered/nsfwjs

Create a voting classifier ensemble for more advanced accuracy

Open

#491 建立於 2021年3月30日

在 GitHub 查看
 (1 留言) (0 反應) (0 負責人)JavaScript (7,090 star) (507 fork)batch import
enhancementhelp wanted

描述

Currently, we have two models. The more accurate one that has type 2 errors, and the slightly less accurate one that has far more type 1 errors

We can combine them with a voting classifier and get higher accuracy.

https://medium.com/@sanchitamangale12/voting-classifier-1be10db6d7a5

Thoughts?

貢獻者指南

技術棧
javascripttensorflow
領域
machine learning
議題類型
feature
難度面向新貢獻者的預計實作難度，1 表示很小改動，5 表示專家級工作。
3
預計時間有經驗貢獻者完成調查、實作、測試並準備 pull request 的粗略時間範圍。
1-2 days
活動狀態議題目前的可參與程度：新鮮、活躍、陳舊、阻塞或等待維護者輸入。
stale
清晰度議題是否清楚說明預期改動、驗收標準和下一步。
needs investigation
前置要求
Understanding of existing NSFW models in the repoBasic knowledge of ensemble methods
新手友善度1-100 的估計分數，表示該議題對首次貢獻者的友善程度。
25
研究方向
Examine the existing models in the repository (likely in /src or /models). Determine how to implement a voting classifier that aggregates predictions from both models. Consider using TensorFlow.js's built in methods or custom logic. Test the ensemble on a dataset to measure accuracy improvement. The issue does not specify implementation details, so further clarification from maintainers may be needed.