PyCon TW 2017 - Day 1

今年 PyCon 終於比起上次聽得懂多了！
看來這兩年來，還是有點進步的 XD

先放上今年的共筆
這幾篇記錄我參加議程的筆記
有些投影片跟共筆就很清楚的，就直接放連結了

議程

• [Keynote] Choices for Smarter AI
• Python 開源軟體考古 - 以Viper為例
• 整合 Slack 與 Docker 搭建 Jupyter 線上程式面試系統
• [Keynote] The State of Python for Education Learning
• Building Microservices in Python 個案分享
  • Slide
  • 共筆
• Understanding Serverless Architecture
  • 同場加映：Don't Reinvent Sandwich (本議程沒供餐 XD)
• Tensorflow & Python: Fault Detection System

[Keynote] Choices for Smarter AI

• 共筆

• Better AI

  • Traditional: 更像人類
  • New: 更好用

Choice[0]: What Language for AI?

• 現場會眾一致通過是 Python (Bias Sampling XD)
• Useful libs
  • xgboost
  • libact: Active Learning

Choice[1]: What Application Should AI Help?

• AI Starts From Problem Solving
  • Motivation
    • Publishable (academia), Profitable
  • Feasibility
    • Modeling, Timeline, Budge
• Big Problems from Big Data
  • Velocity: Evolving data, Evolving problems
  • Volume: Computational Bottleneck
  • Veracity: Modeling with non-textbook data → Noise, Bias

Choice[2]; What Route for AI

Tip: As much human as possible before going to machine

Choice[3]: How to Measure AI Goal?

Tip: Start with reasonable, measurable and prioritized goals for AI

Choice[4]: What Data to (or not to) Use?

• Choice factors for data
  • Utility: Relationship with goal
  • Necessity: Uniqueness to goal
  • Quality: Noise, Freshness
  • Cost

Tip: Start with "minimum viable data"

Choice[5]: What Model to Start?

Linear (Simpler) Model First

Choice[6]: What Improvement Steps to Take?

• Lose Reason
  • Overfitting
  • Misfitting
  • Over-reusing
    • Keep data fresh

Choice[-1]: How to verify and Deploy?

• Human Trust matters
  • Need a baseline to be compared

Misc[0]: No Choice is a Choice

Misc[1]: Learning from Mistake

Misc[2]: ???

Python 開源軟體考古 - 以 Viper 為例

這場很實用，slide 也很清楚
蠻推薦影片出來可以看一下

• slide
• 共筆
• viper-research

從開源專案學習寫 code

讀 code 技巧

降低專案複雜度

• 從早期版本追
  • 如何挑版本？ ( 搭配 tig 服用 )
    • 重大版本號
    • 簡單、可運作之版本 (e.g. viper 的 commit hash: 46a2a)
• 感覺太複雜？
  • 砍！
  • 鎖定特定功能，移除其他雜質
  • 測試，能動就可以

專案程式邏輯架構

模組相依性 → 一直 trace 到沒有 import 專案自己寫的 code

• Tools
  • modulegraph
  • pydegraph (py2degraph)
  • snackfood
• 數據分析
  • e.g. 被用最多的反而不是核心 → 這些程式碼好用、易用
• 走訪專案
  • 建立專案整體架構邏輯
  • 深度走訪
    • 由下往上
    • 仔細閱讀單一程式
  • 廣度走訪
    • 由上往下
    • 解釋特定組合的程式的意義

Conclusion

• 系統化讀 code
• 從 Commit 學習
  • 架構變化
  • Commit Message 規則
  • Branching Model
  • Issue Handling

整合 Slack 與 Docker 搭建 Jupyter 線上程式面試系統

• slide
• pnp-interview
• 共筆

這場最重要的大概就是 slide 第 12 頁 第 11 頁的架構圖

Tools Used

• Flask
  • Python 中最簡單使用的 web framework
  • 做小型 web 應用非常適合
• Docker
  • 容器化
  • 一鍵部署
  • 限制容器耗費的 CPU, GPU
• Slack
  • Integration 很好

Bugs

• Pull Image First
  • So it can be fast
• Try except for any case
  • dockerpy 的雷 xD
• File Permission
  • 要採 docker 坑，這很重要

[Keynote] The State of Python for Education Learning

• slide
• 共筆

這場 Talk 主要講學習、推廣和社群

Carol 有提到幾個學習 Python 很棒的資源
其中我覺得最有用的大概就是pyvideo.org 了
之前回去聽工資管系系友演講，趨勢的學長就有提到看 Talk 是很快的學習方式

另外，Carol 強力推薦今年 PyCon US，Instagram 給的 Talk
Lisa Guo, Hui Ding Keynote PyCon 2017
之後，應該也會找個時間來看一下

Understanding Serverless Architecture

• slide
• 共筆

Serverless

• Function as a service (FaaS)
  • e.g. AWS lambda
• Advantage
  • Don't need to maintain servers
• Disadvantages
  • Functions are allowed to run for only a limited amount of time
  • Heavy workloads cannot be run
  • No control over containers
  • Hard to monitor
  • Hard to scale up
• It's awesome but not the best choice for everyone.

Tensorflow & Python: Fault Detection System

• slide
• 共筆

Fault: An abnormal condition or defect at the component

• Logs
  • Usage of CPU
  • Memory
  • Disk I/O
  • Network Bandwidth
  • System Log
  • Application Log
  • and etc.

Log is also natural language.
The sequence of words and expression is important sequential data.

這場我真的就有點聽不太懂了＠＠