Kaggle 필사

코딩을 공부하는데 있어 직접 코드를 짜는 것도 중요하지만, 다른 사람이 짜놓은 좋은 코드를 따라 쳐보는 것도 좋은 공부가 된다고 한다. 대표적으로 kaggle 사이트 대회의 정답을 따라하는 것이 특히 도움이 된다고 하니 추가로 공부하는 것도 나쁘지 않아보인다.

 

캐글 필사 커리큘럼은 다음과 같다.

 

https://kaggle-kr.tistory.com/32

[이유한님] 캐글 코리아 캐글 스터디 커널 커리큘럼

 

참여방법은 간단하다.

 

커리큘럼 참여 방법

• 커널의 A 부터 Z 까지 똑같이 3번 따라 적기
• 진행 완료된 커리큘럼의 경우 링크 해제

 

Binary classification : Tabular data

1st level. Titanic: Machine Learning from Disaster

• 타이타닉 튜토리얼 1 - Exploratory data analysis, visualization, machine learning
• EDA To Prediction(DieTanic)
• Titanic Top 4% with ensemble modeling
• Introduction to Ensembling/Stacking in Python

2nd level. Porto Seguro’s Safe Driver Prediction

• Data Preparation & Exploration
• Interactive Porto Insights - A Plot.ly Tutorial
• XGBoost CV (LB .284)
• Porto Seguro Exploratory Analysis and Prediction

3rd level. Home Credit Default Risk

• Introduction: Home Credit Default Risk Competition
• Introduction to Manual Feature Engineering
• Stacking Test-Sklearn, XGBoost, CatBoost, LightGBM
• LightGBM 7th place solution

Multi-class classification : Tabular data

1st level. Costa Rican Household Poverty Level Prediction

• A Complete Introduction and Walkthrough
• 3250feats->532 feats using shap[LB: 0.436]
• XGBoost

Binary classification : Image classification

1st level. Statoil/C-CORE Iceberg Classifier Challenge

• Keras Model for Beginners (0.210 on LB)+EDA+R&D
• Transfer Learning with VGG-16 CNN+AUG LB 0.1712
• Submarineering.EVEN BETTER PUBLIC SCORE until now.
• Keras+TF LB 0.18

Multi-class classification : Image classification

1st level. TensorFlow Speech Recognition Challenge

• Speech representation and data exploration
• Light-Weight CNN LB 0.74
• WavCeption V1: a 1-D Inception approach (LB 0.76)

Regression : Tabular data

1st level. New York City Taxi Trip Duration

• Dynamics of New York city - Animation
• EDA + Baseline Model
• Beat the benchmark!

2nd level. Zillow Prize: Zillow’s Home Value Prediction (Zestimate)

• Simple Exploration Notebook - Zillow Prize
• Simple XGBoost Starter (~0.0655)
• Zillow EDA On Missing Values & Multicollinearity
• XGBoost, LightGBM, and OLS and NN

Object segmentation : Deep learning

1st level. 2018 Data Science Bowl

• Teaching notebook for total imaging newbies
• Keras U-Net starter - LB 0.277
• Nuclei Overview to Submission

Natural language processing : classification, regression

1st level. Spooky Author Identification

• Spooky NLP and Topic Modelling tutorial
• Approaching (Almost) Any NLP Problem on Kaggle
• Simple Feature Engg Notebook - Spooky Author

2nd level. Mercari Price Suggestion Challenge

• Mercari Interactive EDA + Topic Modelling
• A simple nn solution with Keras (~0.48611 PL)
• Ridge (LB 0.41943)
• LGB and FM [18th Place - 0.40604]

3rd level. Toxic Comment Classification Challenge

• [For Beginners] Tackling Toxic Using Keras
• Stop the S@#$ - Toxic Comments EDA
• Logistic regression with words and char n-grams
• Classifying multi-label comments (0.9741 lb)

Other dataset : anomaly detection, visualization

1st level. Credit Card Fraud Detection

• In depth skewed data classif. (93% recall acc now)
• Anomaly Detection - Credit Card Fraud Analysis
• Semi-Supervised Anomaly Detection Survey

2nd level. Kaggle Machine Learning & Data Science Survey 2017

• Novice to Grandmaster
• What do Kagglers say about Data Science ?
• PLOTLY TUTORIAL - 1