more datacamp

here are my notes on the first 3 chapters of the datacamp courses i did today (a continuation of experimental design in python and supervised learning with scikit-learn):

chapter 4: advanced insights from experimental complexity

• addressing complexities in experimental data
  • heteroscedasticity - changing variability of a variable across the range of another variable

chapter 1: classification

• machine learning with scikit-learn
  • machine learning - the process in which computers learn to make decisions from data without being explicitly programmed
  • unsupervised learning - the uncovering of hidden patterns from unlabelled data
  • supervised learning - a type of machine learning in which the predicted values are known and the model is built to accurately predict values of previously unseen data
• the classification challenge
  • the 4 steps of classifying labels of unseen data:
    • building a model
    • have the model learn from labelled data we pass to it
    • pass unlabelled data to the model as input
    • model predicts labels of the unseen data
  • labelled data = training data
  • k-nearest neighbours algorithm - used to predict the label of any data point by looking at the k closest labelled data points, and by taking a majority vote
  • measuring model performance
    • accuracy - correct predictions / total observations
• covariate adjustment in experimental design
  • covariates - variables that are related to the outcome variable and can influence its analysis
  • can help in reducing confounding

chapter 2: regression

• introduction to regression
  • (another type of supervised learning) in regression, target variable typically has continuous values
• the basics of linear regression
  • y=ax+b
    • where y is the target
    • x is the single feature in simple linear regression
    • a and b are prarmeters or the coefficients of the model (slope and intercept)
  • when adding more features: y= a1x1 + a2x2 + …. anxn + b
• cross-validation
  • steps of k-fold cross-validation:
    • split dataset into groups / folds
    • set aside first fold as test set
    • fit model on remaining folds
    • predict on test set
    • computer metric of interest
    • repeat with the next folds
  • regularised regression
    • regularisation is used to avoid overfitting in regression
    • penalises large coefficients