Cocolour

Color schemes generator based on machine learning

Development

sudo npm install -g grunt-cli
npm install

Build

grunt build

Watch

grunt watch

Test

sudo npm install -g coffee-script
npm test

FAQ

Why HSL?

Section: HSL与HSV的比较

讨厌喜欢的分离

事实上，似乎喜欢和讨厌的机制是很不一样的，所以他们应该被丢到两个堆中去。

如果直接用单个score输出，正确率非常低，只有三十几。如果用 [喜欢，不喜欢，一般] 输出，大概61% 如果用 [喜欢, 不喜欢]，大概68-69%

https://github.com/zenozeng/cocolour/issues/77

经验

BP神经网络命中率波动非常厉害，而且很低？

可能是数据过拟合了，调整 iterations 和 errorThresh 看看会不会有帮助。

License

Copyright (C) 2014-2015 Zeno Zeng

This program is free software: you can redistribute it and/or modify
it under the terms of the GNU Affero General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.

This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
GNU Affero General Public License for more details.

You should have received a copy of the GNU Affero General Public License
along with this program.  If not, see <http://www.gnu.org/licenses/>.

This program incorporates work covered by the following copyright and permission notices:

jQuery

Copyright 2005, 2014 jQuery Foundation, Inc. and other contributors

Released under the MIT license
color-convert

Copyright (c) 2011 Heather Arthur [email protected]

Released under the MIT license
colors-clustering

Copyright (c) 2014 Zeno Zeng

Released under the MIT license
gene-pool

Copyright (c) 2014 Zeno Zeng

Released under the MIT license
brain

Copyright (c) 2010 Heather Arthur

Released under the MIT license

项目日程

2015-04-06 -- 2015-04-12

add ANN.prototype.toJSON
add ANN.fromJSON

2015-03-30 -- 2015-04-05

force login to use setScore, fixes #123

2015-03-16 -- 2015-03-22

add getError() in ANN.train

2015-03-09 -- 2015-03-15

模拟的举证里增加 math.var(distance, 'biased')
数据增加到 1441 组
将不同人的数据分离进行验证

2015-03-02 -- 2015-03-08

换用 convnetjs 的讨论
关于个人训练模型的讨论
Switch to convnetjs
Try SVM layer and switch back to softmax
Add min distance

2015-03-23 -- 2015-03-01

UI 细节调整、加载速度优化
现在切分训练数据与验证数据时会保持 positive 和 negative 的比例

2015-02-16 -- 2015-02-22

网站 UI 的进一步重构
代码组织方式调整、重构

2015-02-09 -- 2015-02-15

一些前端小调整
重写神经网络预测率测试
- 基于 cluster 模块，使用 process 的 message 通信
- 输出测试结果的 max, min, mean, median 和 var
调整 iterations 和 errorThresh 来避免过拟合问题
- 平均预测率基本可以到 71%
- 而且现在输出结果不再发生大幅度波动了
See also: https://github.com/zenozeng/cocolour/issues/92
移除色相、亮度、饱和度方差
- 平均预测率现在可达 72%
See also: https://github.com/zenozeng/cocolour/issues/78
输出讨厌和喜欢各自的预测率
- 目前喜欢可达 80%，讨厌仅为 56%
See also: https://github.com/zenozeng/cocolour/issues/90
尝试给估价函数增加一个 bias
```
bias = 0.38
(positive - bias) * 1 / bias
```
对喜欢的平均预测率现在可达 93%，虽然对讨厌的平均预测率现在降到了 39%，综合判断率下降

See also: https://github.com/zenozeng/cocolour/issues/93
网站 UI 调整

2014-10-20 -- 2014-11-09

确定神经网络库的选择为 Brain
确定输入格式为一个 HSL 矩阵的 flatten: H1 S1 L1 H2 S2 L2 H3 S3 L3 H4 S4 L4 H5 S5 L5

See also: https://github.com/zenozeng/cocolour/issues/72
一次概念验证性测试
```
Length:  164
Match Cound:  106
Unmatch Cound:  58
Rate(%):  64.63414634146342
```
具体的测试详情：https://github.com/zenozeng/cocolour/issues/76
确定输出格式

似乎喜欢和讨厌的机制是很不一样的，所以他们应该被丢到两个堆中去。

如果直接用单个score输出，正确率非常低，只有30%-40% 如果用 [喜欢，不喜欢，一般] 输出，大概60-61%，以及一般这一档的数据非常少，不怎么可靠。如果用 [喜欢, 不喜欢]，大概60-68%

具体的测试详情：https://github.com/zenozeng/cocolour/issues/77
增加数据到 813 组
数据分组成 train 和 verify 组的时候引入随机性

https://github.com/zenozeng/cocolour/issues/81

这个 Issue 会导致之前的测定结果存在一定的偏差
增加数据到 1378 组
调整 learningRate 到 0.1

似乎结果的稳定性提升了一些、正确率也提升了一些
基于 master-slave 的多进程结果验证

充分利用多核性能
尝试引入色相方差、饱和度方差、明度方差
调整学习速率到 0.05

2014-10-13 -- 2014-10-19

Fix bugs in UI
Script for fetching all color schemes in database
500+ more color schemes
Normalize colors

2014-10-06 -- 2014-10-12

引入 AVOS Cloud SDK
user.signup, user.login, user.logout & user.passwordReset
DB: Class Scheme
ACL for Scheme
Log heart and trash

2014-08-18 -- 2014-08-24

神经网络库的选择讨论

See also: https://github.com/zenozeng/cocolour/issues/53
界面增加动画
尝试引入遗传算法，以便在更短时间获得更好结果
构建遗传算法库

https://github.com/zenozeng/gene-pool
Move static/font-awesome to cdn.staticfile.org
各家 Baas 服务商的比较，打算使用 avoscloud

https://cn.avoscloud.com/docs/js_guide.html
遗传算法可视化

http://zenozeng.github.io/gene-pool/demo/
引入遗传算法 ([email protected])

2014-08-11 -- 2014-08-17

Go back using CoffeeScript

See also https://github.com/zenozeng/cocolour/issues/37
update header, fixes #41
聚类算法可视化 (D3)

https://github.com/zenozeng/colors-clustering-visualization

2014-06-23 -- 2014-06-29

About whether to use DBaaS or Baas

See also https://github.com/zenozeng/cocolour/issues/46

See also https://github.com/zenozeng/cocolour/issues/44
Consider using Genetic Algorithms

See also https://github.com/zenozeng/cocolour/issues/49

2014-06-16 -- 2014-06-22

Consider using Web Worker
New Arch Design (ClojureScript for Pure Calculation & CoffeeScript for UI and Communication)

2014-06-09 -- 2014-06-15

关于应用容器化的构想，及相关服务提供商的比较

Linode + Ubuntu + Docker / DigitalOcean + Ubuntu + Docker / Stackdock / Tutum

2014-06-02 -- 2014-06-08

New UI Design for colors clustering (in Zeno's loose notes 2014-06-08)

2014-05-12 -- 2014-05-18

Simple JSON based user system
Simple loging system for replaying requests later

2014-05-05 -- 2014-05-11

New name: cocolour
New domain: cocolour.com
Deploy on Github Pages
Move clustering/ to new repo: zenozeng/colors-clustering
Use seeds from CSS Color Module Level 3
Use CIEDE2000 for calc color difference
Add RGBA Support for Colors Clustering
Switch to CIE76 for perfermence

see https://github.com/zenozeng/colors-clustering/issues/7
Add nodejs support for colors-clustering
Npm publish zenozeng/colors-clustering
Rewrite cocolour using zenozeng/colors-clustering
New UI for cocolour
Use Grunt as task runner
UI for 1920 * 1080
New Repo: cocolour-server