Skip to content
YIZE
P.005 / PROJECT2025

智能集成机器学习建模与可视化分析平台Integrated Machine-Learning Modelling and Visualisation Platform

Integrated Machine-Learning Modelling and Visualisation Platform智能集成机器学习建模与可视化分析平台

把常用的建模与可视化分析收进一个平台的登记软件(软著 2025SR1800232),是育种家平台三项软著组件之一。

A registered platform (软件著作权 2025SR1800232) that wraps standard modelling and visual analysis into one place — one of the three copyrighted components underneath the breeding platform.

角色Role
开发成员Developer
类型Type
软件Software
年份Year
2025
  • Python
  • Machine learning
  • Visualisation
Platform Screenshot
P.005-A
Platform Screenshot · placeholder
目录Contents

它是什么

一个把「建模」和「看清模型」放在同一个界面里的桌面平台:数据导入、降维、集成建模、评价、解释,各自是一个可以单独走完的步骤,也可以连起来当一条流程走。登记号 2025SR1800232,著作权人为中国农业大学,是育种家平台之下的三项软著组件之一。

为什么这件事需要单独做一个软件

做生物学或农学的分析的人,面对的往往是同一个循环:把表导进来、降个维、试几个模型、看哪个变量有用、再解释给合作者听。这个循环里每一步都有成熟的库,但每一步的环境都不一样——一边是 Python 的依赖,一边是画图的脚本,中间还夹着若干次手工转格式。做过几轮之后,重复劳动的比例会比分析本身还高。

平台的定位就是把这条循环收进一个界面,而且是把「解释」也收进去,而不是停在给出准确率。

功能

  • 数据导入:CSV 与 Excel。
  • 降维:主成分分析。
  • 集成建模:以 XGBoost 为代表的梯度提升,以及 Bagging / Boosting 两类集成策略。
  • 评价:ROC 等分类评价指标。
  • 解释:SHAP 一类的特征贡献分析,用于回答「哪个变量在起作用」——这是 SNP 数据上最常被问到、也最容易被跳过的一步。

技术选型

  • 前端:Vue 3 + Vite + Naive UI,外面套 Tauri 作为桌面外壳。
  • 后端:Python + FastAPI。
  • 通信:前端与后端之间走 WebSocket,而不是一问一答的 HTTP。

最后一条是这几个选择里唯一有解释价值的:模型训练是一件会持续几十秒到几分钟的事,如果界面只能等一个响应,用户就不知道它是在算、在卡住,还是已经死了。WebSocket 让进度能推回来,界面在整个过程中都是活的。

跨平台:Windows、Linux 与 macOS 都能跑——这也是选 Tauri 而不是绑定单一系统的桌面框架的原因,后端只要有 Python 就行,不依赖 GPU。

这一页不写的

界面截图。软件是从中国版权保护中心登记的作品,页面上标注登记号即可核验;把界面图贴出来并不增加可核对性,反而容易变成装饰。

目录Contents

What it is

A desktop platform that puts modelling and understanding-the-model in the same window: data import, dimensionality reduction, ensemble modelling, evaluation and interpretation, each usable as a step on its own and connectable into one run. Registered as 2025SR1800232, rights holder 中国农业大学, and one of the three copyrighted components underneath the breeding platform (P.001).

Why this needed its own program

Anyone doing analysis in biology or agriculture meets the same loop: import the table, reduce its dimensions, try a few models, see which variables matter, then explain it to a collaborator. There is a mature library for every step of that loop — but not one environment for all of them. Python dependencies on one side, plotting scripts on another, and several rounds of hand-converting formats in between. After a few passes, the repeated labour outweighs the analysis.

The platform's aim is to hold that loop in one interface, including the explanation step, rather than stopping at a score.

What it does

  • Import — CSV and Excel.
  • Reduction — principal component analysis.
  • Ensemble modelling — gradient boosting with XGBoost, and bagging and boosting as ensemble strategies.
  • Evaluation — classification metrics including ROC.
  • Interpretation — feature-contribution analysis of the SHAP kind, to answer which variable is doing the work. On SNP data that is the question most often asked and most often skipped.

Why the stack looks like this

  • Front end: Vue 3 + Vite + Naive UI, wrapped in Tauri as the desktop shell.
  • Back end: Python + FastAPI.
  • Transport: WebSocket between the two, not request/response HTTP.

The last one is the only choice here worth explaining. Training a model takes anywhere from tens of seconds to minutes; if the interface can only await one response, the person watching cannot tell whether it is computing, stuck, or dead. A WebSocket lets progress be pushed back, which keeps the interface alive for the whole run.

Cross-platform: Windows, Linux and macOS, which is why the shell is Tauri rather than a framework tied to one system — the back end needs nothing but Python, and no GPU.

What is not on this page

Screenshots. The software is registered with the Copyright Protection Center of China, and its registration number above is what makes it checkable; a picture of the interface adds nothing to that and tends to become decoration.