mofhu/didi-transition

Didi Transition Predict Contest. Copyright Org Lab 2016

didi-transition's Issues

QC outline

Proposal

I propose QC outline, which 设计本周项目轮廓, 并设立关键质控指标.

Introduction

Need an outline for this week. checklist and QC benchmark method also needed.

Method

Major part of this proposal and implementing principle.

Implementing details

Detailed checklist for implementing:

• 初步清洗数据 #4
• 预测脚本 #5
• 简单预测模型与QC (benchmark: local and online)
• 搭建稍复杂模型

QC & benchmark

• Wed: QC online result (better than all 1)
• Fri: QC online result (复杂模型的效果应有进一步提升)

Assignee

@mofhu

---

Copyright 2016 Organization Laboratory

First test outline

Proposal

I propose first tests, which first test data and decide future outline.

Introduction

需要先质控一下项目, 用最简代码尝试提交 (2-3 次). 决定是否投入更多精力参赛(或放弃).

Method

设计三次提交: 全0; 均值; 最简预测.

Implementing details

Branch to work at: master/dev.

Detailed checklist for implementing:

• 洗数据
• 提交均值
  • 洗数据
  • 提交
• 提交最简预测值
  • 设计本地测试集
  • 最简预测模型(工程时间 <1d)
  • 本地 benchmark
  • 提交
• QC position

Assignee

@mofhu

---

Copyright 2016 Organization Laboratory

Pivottable

Proposal

I propose pivottable, which 从数据透视中了解数据情况, 为预测做准备.

Introduction

数据理解和清洗应占据大部分工程时间; 并且目前处于项目初期, 数据清洗所占时间比例应更高.

Method

Major part of this proposal and implementing principle.

Implementing details

Branch to work at: branch name.

Detailed checklist for implementing:

• pivottable with all order count and NaN order count
• watch data outline (visualization) for some districts (work, living, etc.)
• try predict (MVP) with these new data

Assignee

@mofhu

---

Copyright 2016 Organization Laboratory

Submissions

Benchmark

Collection of online submissions

Testing parameters

Online submission (http://research.xiaojukeji.com/competition/upload.action?competitionId=DiTech2016)

Actual behavior

date, commit, method abstract, result

---

Copyright 2016 Organization Laboratory

Scoring script

Proposal

I propose Scoring script, which 本地给CV/测试集打分.

Introduction

Notes of motivation.

Method

Major part of this proposal and implementing principle.

Implementing details

Branch to work at: dev.

Detailed checklist for implementing:

• 读取标答和预测集
• 按照公式给出答案
• 测试 (用已有测试集其它时间的数据)

Assignee

@mofhu

---

Copyright 2016 Organization Laboratory