datacamp studies

chapter 3: bagging and random forests

• bagging
  • also called bootstrap aggregation
  • reduces variance of individual models in ensemble
• out of bag evaluation
  • out of bag (oob) instances - on average 37% of training instances sampled for each model

chapter 4: boosting

• adaboost
  • boosting - ensemble method that combines several weak learner to form a strong learner
  • weak learner - model doing just slightly better than random guessing
  • adaboost stands for adaptive boosting - pays more attentiont to instances wrongly predicted by predecessor
• gradient boosting (gb)
  • in gradient boosting, each predictor corrects its predecessor’s error
• stochastic gradient boosting (sgb)
  • trees are trained on a random subset of rows of training data
  • sampled instances are without replacement

language trilogy

started my maru hiragana drills at 12.47 and finished them in 3 minutes (100 out of 105 hiragana characters recognised)

13.19 - finished chapter 10 busuu japanese

13.35 - finished half of chapter 5 busuu french

13.36 - finished half of chapter 5 busuu german