visualdl / visualdl Goto Github PK

View Code? Open in Web Editor NEW

13.0 10.0 1.0 201 KB

License: Apache License 2.0

CMake 6.03% Python 19.27% C++ 36.85% Shell 0.74% JavaScript 32.88% CSS 3.88% HTML 0.35%

visualdl's Introduction

VisualDL

visualdl's People

Stargazers

Watchers

Forkers

tttangtang

visualdl's Issues

Warn-Up!

DNN 中可视化场景（支持难度由低到高）

常规训练信息概览

Train/Test error plot
准确率曲线
网络配置可视化
任意 tensor 可视化

典型案例

图像可视化（矩阵解析为图片，可用于CNN可视化）
Embedding 可视化

初版包含功能

版本发布日期规划

embed backend code autostyle toolchain into pre-commit

languages to support:

Python
c++

usage:

pre-commit install
pre-commit or git commit will invoke the toolchain

backend primary API

后端-基本设计文档

基本模块设计

Fix Misspellings

Need to fix some misspellings in the code:

toogle, sery, visaul, filted

https://github.com/VisualDL/VisualDL/blob/develop/frontend/src/common/ui/Charts/chart.san#L191
https://github.com/VisualDL/VisualDL/blob/develop/frontend/src/common/ui/ExpandPanel.san#L31
https://github.com/VisualDL/VisualDL/blob/develop/frontend/src/common/ui/ExpandPanel.san#L33
https://github.com/VisualDL/VisualDL/blob/develop/frontend/src/common/ui/ExpandPanel.san#L40
https://github.com/VisualDL/VisualDL/blob/develop/frontend/src/scalars/Scalars.san#L129

init cmake system

Cmake should download dependency automatically.
each module should have a CMakeLists.txt

the python package build workflow

可以把python包封装的工作流预先建立起来

The first issues of the VisualDL project

The following plan will be discussed here.

确定后端protobuf

primary server

Looking forward to !

define get tablet interface

design of backend log format

可视化的功能包括

Graphs, 用一个 graph 可视化模型相关的整个神经网络，包括每层网络中数据的输入输出、涉及的参数等。
Scalars, 模型训练相关的所有标量信息。
Images, 识别相关的图像信息或者模型可解释性等。
Histograms
Distributions

frontend&&backend interface design

overall design of backend

Do Research on TensorBoard API

Try to figure out how Scalar and Image data are retrieved via API and the meanings of parameters and data formats

need to refactor Storage to support data in memory or disk

Create virtualenv so that versions are consistent

For example, we are using

TensorBoard 0.4.0rc3

as standard but some developers might use other version.

storage.pb.h cannot be generated automatically

VisualDL is Great!

Write Demo / Tutorial / Survey under Docs for VisualDL

By Jan 2nd, Scalar and Image should be ready. Write corresponding demo / tutorial / Survey for these 2 features.