🏆 Tiêu chuẩn — Capstone Project Kickoff

Các tiêu chuẩn giúp bạn kickoff capstone project đúng quy trình, đúng cấu trúc, và reproducible — TDSP cho project lifecycle, Git Workflow cho version control, Reproducible Analysis cho tính minh bạch.

Tổng quan tiêu chuẩn buổi 18

Buổi 18 bắt đầu Capstone Project — từ chọn đề tài, data collection, đến setup repository. Để project đạt chuẩn portfolio-ready, cần tuân theo 3 tiêu chuẩn từ industry:

TDSP (Microsoft) — Team Data Science Process → quy trình project lifecycle cho data science/analytics
Git Workflow Standards — Conventional Commits + Branching Strategy → version control chuyên nghiệp
Reproducible Analysis — Tiêu chuẩn reproducibility cho analytical work → người khác có thể verify kết quả

📋 Danh sách tiêu chuẩn liên quan

#	Tiêu chuẩn	Tổ chức / Tác giả	Áp dụng cho Buổi 18
1	TDSP	Microsoft (2016-present)	Project lifecycle, folder structure, documentation
2	Conventional Commits + Git Flow	Git community / Vincent Driessen	Commit messages, branching, version control workflow
3	Reproducible Analysis	ROpenSci / Turing Way / Wilson et al.	Environment management, documentation, code practices

1️⃣ TDSP — Team Data Science Process (Microsoft)

Giới thiệu

TDSP (Team Data Science Process) — framework do Microsoft phát triển (2016, cập nhật liên tục), cung cấp structured lifecycle cho data science và analytics projects. TDSP khác CRISP-DM (Buổi 17) ở chỗ: TDSP tập trung vào team collaboration, standardized project structure, và Git integration — rất phù hợp cho capstone portfolio project.

TDSP gồm 5 lifecycle stages — iterative, data-driven:

mermaid

flowchart TD
    A["1️⃣ Business<br/>Understanding"] --> B["2️⃣ Data Acquisition<br/>& Understanding"]
    B --> C["3️⃣ Modeling"]
    C --> D["4️⃣ Deployment"]
    D --> E["5️⃣ Customer<br/>Acceptance"]
    E -->|"Iterate"| A
    B -->|"Need clarity"| A
    C -->|"Need more data"| B

5 Lifecycle Stages

Stage	Mục đích	Key Activities	Output cho Capstone
1. Business Understanding	Xác định mục tiêu dự án, metrics, stakeholders	Define objectives, identify data sources, create project plan	Project Brief document
2. Data Acquisition & Understanding	Thu thập data, EDA, đánh giá quality	Ingest data, explore, clean, feature engineering	Data Audit Report, Clean Dataset
3. Modeling	Xây dựng và đánh giá models/analyses	Feature selection, model training, evaluation	Analysis Notebooks, Model Results
4. Deployment	Triển khai results vào production/portfolio	Dashboard, report, API, presentation	Dashboard + Presentation
5. Customer Acceptance	Stakeholder review, handoff	Present findings, document handoff, feedback	Final Presentation + Peer Review

TDSP Standardized Project Structure

TDSP define cấu trúc folder chuẩn cho mọi data project — áp dụng trực tiếp cho capstone:

📁 TDSP Project Structure (adapted for Capstone)
├── 📄 README.md                    ← Project overview (charter)
├── 📄 .gitignore
├── 📁 docs/
│   ├── 📄 project_charter.md      ← Business understanding
│   ├── 📄 data_dictionary.md      ← Data documentation
│   ├── 📄 data_report.md          ← Data quality & EDA summary
│   └── 📄 final_report.md         ← Analysis findings & recs
├── 📁 data/
│   ├── 📁 raw/                    ← Original, immutable data
│   ├── 📁 processed/              ← Cleaned, transformed data
│   └── 📁 external/               ← Third-party data sources
├── 📁 notebooks/                   ← Jupyter notebooks (numbered)
├── 📁 src/                         ← Source code (reusable)
│   ├── 📄 data_processing.py
│   ├── 📄 feature_engineering.py
│   └── 📄 visualization.py
├── 📁 reports/                     ← Generated reports, slides
│   ├── 📁 figures/                ← Saved plots/charts
│   └── 📄 presentation.pdf
└── 📄 requirements.txt            ← Python dependencies

TDSP Project Charter Template

Microsoft recommend mỗi project bắt đầu bằng Project Charter — document 1-2 trang:

PROJECT CHARTER
===============
1. Project Title: [Tên phù hợp, descriptive]
2. Business Background: [Tại sao project này? Stakeholder nào?]
3. Scope:
   - In scope: [Phân tích nào sẽ thực hiện?]
   - Out of scope: [KHÔNG làm gì?]
4. Metrics: [Success metrics — accuracy, business KPI impact]
5. Architecture:
   - Data: [Sources, size, format]
   - Tools: [Python, SQL, BI tool]
   - Deliverables: [Notebook, dashboard, presentation]
6. Timeline: [Milestones theo tuần]
7. Team: [Roles — trong capstone: bạn = all roles]
8. Communication: [Update schedule, feedback loops]

TDSP Checklist cho Capstone

STAGE 1 — BUSINESS UNDERSTANDING:
✅ Project Charter hoàn thành
✅ 3-5 business questions cụ thể
✅ Success metrics defined
✅ Scope clearly bounded (in vs out)

STAGE 2 — DATA ACQUISITION & UNDERSTANDING:
✅ Dataset sourced + downloaded
✅ Data dictionary created
✅ Data quality report (audit)
✅ EDA completed, patterns documented

STAGE 3 — MODELING:
✅ Analysis/models aligned với business questions
✅ Results validated + interpreted
✅ Limitations documented

STAGE 4 — DEPLOYMENT:
✅ Dashboard deployed (Tableau Public / Power BI)
✅ GitHub repo organized per TDSP structure
✅ README updated with findings

STAGE 5 — CUSTOMER ACCEPTANCE:
✅ Presentation rehearsed
✅ Peer review completed
✅ Feedback incorporated

2️⃣ Git Workflow Standards — Conventional Commits + Branching

Giới thiệu

Git Workflow Standards bao gồm 2 conventions quan trọng cho DA projects:

Conventional Commits — chuẩn viết commit messages nhất quán, machine-readable
Git Flow / GitHub Flow — branching strategy cho project development

Dù capstone là solo project, tuân theo Git standards cho thấy bạn work professionally — hiring managers kiểm tra commit history.

Conventional Commits Specification

Format:

<type>[optional scope]: <description>

[optional body]

[optional footer]

Types phổ biến cho DA project:

Type	Mô tả	Ví dụ
`feat`	New feature / analysis	`feat: add RFM customer segmentation`
`fix`	Sửa bug / error	`fix: correct revenue calculation formula`
`data`	Data-related changes	`data: add cleaned dataset v2`
`docs`	Documentation only	`docs: update README with key findings`
`viz`	Visualization changes	`viz: create monthly revenue trend chart`
`refactor`	Code restructure	`refactor: extract helper functions to utils.py`
`style`	Formatting, no logic change	`style: consistent chart color palette`
`test`	Adding/updating tests	`test: add data validation checks`
`chore`	Maintenance tasks	`chore: update .gitignore, add requirements.txt`

Commit Message Best Practices

# ❌ BAD commits (non-informative):
"update"
"fix stuff"
"WIP"
"."
"final version"
"final final version v2 REAL FINAL"

# ✅ GOOD commits (Conventional Commits):
"feat: add exploratory data analysis notebook with distribution plots"
"data: clean missing values in customer_age using median imputation"
"viz: create Tableau dashboard with 4 tabs — overview, segments, delivery, revenue"
"docs: add data dictionary with descriptions for all 15 columns"
"fix: remove duplicate orders (2,341 rows) from analysis dataset"

Branching Strategy cho DA

Simplified GitHub Flow (recommended cho capstone):

mermaid

gitGraph
    commit id: "init"
    commit id: "docs: README"
    branch eda
    commit id: "feat: EDA notebook"
    commit id: "viz: distribution plots"
    checkout main
    merge eda id: "merge: EDA complete"
    branch analysis
    commit id: "feat: segmentation"
    commit id: "feat: delivery analysis"
    checkout main
    merge analysis id: "merge: Analysis done"
    commit id: "docs: final README"

Branch	Purpose	Khi nào dùng
`main`	Stable, presentable version	Chỉ merge code đã test
`eda`	Exploratory analysis	Data cleaning + EDA phase
`analysis`	Deep analysis / modeling	Khi answer specific questions
`dashboard`	BI development	Dashboard build phase
`docs`	Documentation updates	README, reports

Git Workflow Checklist

DAILY WORKFLOW:
✅ git status — check trạng thái trước khi làm
✅ git add <files> — stage files cụ thể (KHÔNG git add . mọi lúc)
✅ git commit -m "type: description" — commit nhỏ, thường xuyên
✅ git push — push lên remote mỗi cuối session
✅ git log --oneline — review lịch sử commits

PROJECT WORKFLOW:
✅ .gitignore setup TRƯỚC khi commit đầu tiên
✅ Branch cho mỗi major phase (eda, analysis, dashboard)
✅ Merge vào main khi phase hoàn thành
✅ Tag releases: v0.1 (data ready), v0.5 (analysis done), v1.0 (complete)
✅ README update mỗi khi có milestone mới

3️⃣ Reproducible Analysis — Standards & Practices

Giới thiệu

Reproducible Analysis — nguyên tắc rằng bất kỳ ai cũng có thể chạy lại analysis của bạn và nhận được CÙNG kết quả. Đây là tiêu chuẩn cốt lõi trong scientific research, và ngày càng quan trọng trong DA — đặc biệt khi analysis ảnh hưởng business decisions.

Theo The Turing Way (Alan Turing Institute, 2024): "Reproducibility is the minimum standard for credible research and analysis. If it can't be reproduced, it shouldn't be trusted."

Cho capstone project, reproducibility có nghĩa: nếu hiring manager clone repo và chạy code, kết quả phải giống hệt output bạn trình bày.

5 Pillars of Reproducible Analysis

mermaid

flowchart TD
    A["🔄 Reproducible Analysis"] --> B["1️⃣ Environment<br/>Management"]
    A --> C["2️⃣ Data<br/>Management"]
    A --> D["3️⃣ Code<br/>Practices"]
    A --> E["4️⃣ Documentation"]
    A --> F["5️⃣ Version<br/>Control"]

Pillar	Principle	Implementation cho Capstone
1. Environment	Pin dependencies, isolate environment	`requirements.txt` với versions, `venv` hoặc `conda env`
2. Data	Raw data immutable, transformations scripted	`data/raw/` (never modify), `data/processed/` (generated by code)
3. Code	No manual steps, everything in scripts/notebooks	Notebooks chạy sequential (01 → 02 → 03), no "run this cell first" chaos
4. Documentation	Assumptions, decisions, limitations documented	Markdown cells, data dictionary, README
5. Version Control	Track changes, meaningful history	Git commits, tags, branching

Environment Management

bash

# === Tạo virtual environment ===
python -m venv venv
source venv/bin/activate  # Mac/Linux
.\venv\Scripts\activate   # Windows

# === Install packages ===
pip install pandas==2.1.4 numpy==1.26.2 matplotlib==3.8.2 seaborn==0.13.0 scikit-learn==1.3.2

# === Freeze requirements ===
pip freeze > requirements.txt

# === Reproduce trên máy khác ===
pip install -r requirements.txt

requirements.txt mẫu cho capstone:

pandas==2.1.4
numpy==1.26.2
matplotlib==3.8.2
seaborn==0.13.0
scikit-learn==1.3.2
jupyter==1.0.0
openpyxl==3.1.2

Code Practices for Reproducibility

Practice	Tại sao?	Ví dụ
Set random seed	Đảm bảo kết quả giống nhau mỗi lần chạy	`np.random.seed(42)`
Relative paths	Code chạy trên máy khác	`pd.read_csv('data/raw/data.csv')` NOT `C:\Users\...`
No hardcoded values	Dễ thay đổi + transparent	`THRESHOLD = 0.05` ở đầu notebook
Functions, not copy-paste	DRY principle, maintainable	`def clean_data(df): ...`
Sequential notebooks	Rõ execution order	`01_cleaning.ipynb → 02_eda.ipynb → 03_analysis.ipynb`
Output regeneratable	Charts/reports tạo từ code	`plt.savefig('reports/figures/chart1.png')`

Reproducibility Checklist cho Capstone

╔═══════════════════════════════════════════════════════════════╗
║          REPRODUCIBILITY CHECKLIST — CAPSTONE                 ║
╠═══════════════════════════════════════════════════════════════╣
║  ENVIRONMENT:                                                 ║
║  ☐ requirements.txt với pinned versions                       ║
║  ☐ Python version documented (≥ 3.9)                          ║
║  ☐ Virtual environment instructions trong README              ║
║                                                               ║
║  DATA:                                                        ║
║  ☐ Raw data unmodified (hoặc download link)                   ║
║  ☐ Data processing fully scripted (no manual Excel edits)     ║
║  ☐ Data dictionary (column descriptions)                      ║
║                                                               ║
║  CODE:                                                        ║
║  ☐ Random seed set                                            ║
║  ☐ Relative paths only                                        ║
║  ☐ Notebooks numbered + sequential                            ║
║  ☐ No "magic numbers" (constants at top)                      ║
║  ☐ Functions for repeated logic                               ║
║                                                               ║
║  DOCUMENTATION:                                               ║
║  ☐ README "How to Run" section                                ║
║  ☐ Assumptions documented                                     ║
║  ☐ Limitations acknowledged                                   ║
║                                                               ║
║  VERSION CONTROL:                                             ║
║  ☐ .gitignore comprehensive                                   ║
║  ☐ Meaningful commit messages                                 ║
║  ☐ No secrets/keys in repo                                    ║
╚═══════════════════════════════════════════════════════════════╝

📊 So sánh 3 tiêu chuẩn

Aspect	TDSP	Git Workflow	Reproducible Analysis
Focus	Project lifecycle & structure	Version control & collaboration	Correctness & verifiability
Origin	Microsoft Azure ML team	Open-source Git community	Scientific research community
Key benefit for capstone	Professional folder structure	Clean commit history impresses hiring managers	Hiring manager can verify your work
Most important element	Project Charter + TDSP folder structure	Conventional Commits + consistent pushing	requirements.txt + relative paths + seed
Time investment	30 min setup	5 min per commit session	15 min initial setup
ROI	Portfolio looks professional	Evidence of consistent work habits	Builds trust in your results

🔑 Tổng kết

#	Tiêu chuẩn	Key Action cho Capstone
1	TDSP	Tạo Project Charter + folder structure chuẩn TDSP
2	Git Workflow	Conventional Commits, commit nhỏ + thường xuyên, branch per phase
3	Reproducible Analysis	requirements.txt + seed + relative paths + sequential notebooks

💡 Tại sao tiêu chuẩn quan trọng cho portfolio?

Hiring managers không chỉ đánh giá kết quả — họ đánh giá quy trình. Capstone project tuân theo TDSP, Git workflow, và reproducibility standards cho thấy bạn work like a professional, không chỉ biết code.

🏆 Tiêu chuẩn — Capstone Project Kickoff ​

Tổng quan tiêu chuẩn buổi 18 ​

📋 Danh sách tiêu chuẩn liên quan ​

1️⃣ TDSP — Team Data Science Process (Microsoft) ​

Giới thiệu ​

5 Lifecycle Stages ​

TDSP Standardized Project Structure ​

TDSP Project Charter Template ​

TDSP Checklist cho Capstone ​

2️⃣ Git Workflow Standards — Conventional Commits + Branching ​

Giới thiệu ​

Conventional Commits Specification ​

Commit Message Best Practices ​

Branching Strategy cho DA ​

Git Workflow Checklist ​

3️⃣ Reproducible Analysis — Standards & Practices ​

Giới thiệu ​

5 Pillars of Reproducible Analysis ​

Environment Management ​

Code Practices for Reproducibility ​

Reproducibility Checklist cho Capstone ​

📊 So sánh 3 tiêu chuẩn ​

🔑 Tổng kết ​

🏆 Tiêu chuẩn — Capstone Project Kickoff

Tổng quan tiêu chuẩn buổi 18

📋 Danh sách tiêu chuẩn liên quan

1️⃣ TDSP — Team Data Science Process (Microsoft)

Giới thiệu

5 Lifecycle Stages

TDSP Standardized Project Structure

TDSP Project Charter Template

TDSP Checklist cho Capstone

2️⃣ Git Workflow Standards — Conventional Commits + Branching

Giới thiệu

Conventional Commits Specification

Commit Message Best Practices

Branching Strategy cho DA

Git Workflow Checklist

3️⃣ Reproducible Analysis — Standards & Practices

Giới thiệu

5 Pillars of Reproducible Analysis

Environment Management

Code Practices for Reproducibility

Reproducibility Checklist cho Capstone

📊 So sánh 3 tiêu chuẩn

🔑 Tổng kết