id-shiv / iutils

Modules for problem solving with Machine Learning

iutils

Modules for problem solving with Machine Learning

Approach

Data terminologies

• Samples: Each row of data.
• Features: Columns headers for each row.
• Targets: Assigned labels for each row.

Supervised Vs Un-supervised

• Supervised: Data with Target variables.
  • Classification: Predicting a class from discrete data.
    • Evaluation Metrics:
      • AUC.
      • Accuracy.
      • Logloss.
      • F1 (Precision / Re-call).
  • Regression: Redicting a value from continuous data.
    • Evaluation Metrics:
      • MAE.
      • MSE.
      • R2.
• Un-supervised: Data without Target variables.

Types of dataset

• Classification problem with 1 sample having 1 class and there are only 2 classes.
• Classification problem with 1 samplae having 1 class and there are more than 2 classes.
• Regression problem predicting of only one value.
• Regression problem predicting many values.
• Classification problem with 1 sample belonging to more than 1 class.

Explore data

• View spread of classes using pandas value_counts

Handling categorical data

• No processing.
• Frequency encoding.
• One-hot encoding.
• Label encoding.

Model

• Scaled.
  • e.g. Linear Regression.
• Non-scaled.
  • XGBoost

Cross Validation

• KFold.
• Stratified KFold (Only for Regression problem).
• Leave one out.
• Leave one group out.