100 Days Challenge Day 14 - Datasets and Feature Engineering

100 Days Challenge - Day 14

Feature Engineering

Pic Credits: How to Win A Data Science Competition Course on Youtube

 

Learnt feature engineering methods from Krish Naik's Youtube channel and videos from How to Win a Data Science Competition Course on Youtube

Topics covered:

• Continuous, Discrete (Numeric and Categorical) and Date-Time Variables
• Loading a random sample with specified features from a csv data file
• Plotting histograms
• Converting date and time object to Pandas date-and-time datatype  

• Feature Generation
• One-Hot Encoding (For Non-Tree based models)

• Scaling and Regularization (for Non Tree-based models)
• Scikit-learn scalers (MinMaxScaler and StandardScaler)
• Handling Outliers using Winsorization and Rank Transformation(Only for Linear Models, KNN and Neural Networks)
• Log Transform and Raising to Power<1 (For neural Networks)
• Generating new features which are functions of existing features

• Categorical and Ordinal Features
• Label Encoding (Works best with Trees)
• Frequency Encoding
• Feature Generation using Categorical Features (Works best for Linear Models and KNN)

Sources:

• Feature Engineering in Python- What are continuous numerical variables?
• Feature Engineering in Python- What are discrete numerical variables?
• Feature Engineering in python-What are date time variable?
• 9 - Feature Engineering Overview | How to Win a Data Science Competition: Learn from Top Kagglers
• 10 - Numeric Features | How to Win a Data Science Competition: Learn from Top Kagglers
• 11 - Categorical and Ordinal Features | How to Win a Data Science Competition: Learn from Top Kagglers