Skip to content

Latest commit

 

History

History
79 lines (61 loc) · 3.45 KB

README.md

File metadata and controls

79 lines (61 loc) · 3.45 KB

Kaggle > Competitions

Titanic - Machine Learning from Disaster

https://www.kaggle.com/competitions/titanic

<List>

R : Random Forest (2023.12.02)

  • 1st attempt to apply the Random Forest method in R
    • Pre-processing : Performed label encoding for Sex Embarked 
      df$Sex <- as.numeric(factor(df$Sex), level=levels(df$Sex))
df$Embarked <- as.numeric(factor(df$Embarked), level=levels(df$Embarked))
    • Utilized the randomForest() function without specific options 
      md <- randomForest(Survived ~ ., data = df_train)
pred <- predict(md, newdata = df_valid, type = "class")
  • Performance Scores (Accuracy) : Not so different from the previous trials with HGB
    • Test (train.csv) : 0.8379888
    • Submission : 0.76794
  • Kaggle Code : Random Forest in Titanic (Version 6)
    • Titanic_RandomForest.r is executable in a local environment

Python : HGB 1.2 (2022.07.31)

  • More advanced HGB(Histogram-based Gradient Boosting)
    • Convert Pclass as a categorical variable additionally 
      df = pd.get_dummies(df, columns=["Pclass", "Embarked", "Sex"])
    • Change the parameter max_iter value from 1000 to 3000 
      hgb = HistGradientBoostingClassifier(max_leaf_nodes=5, learning_rate=0.01, max_iter=3000, random_state=604)
  • Performance Scores (Accuracy)
    • Test (train.csv) : 0.8547486033519553
    • Submission : 0.74641 (rather stepped back??)
  • Kaggle Code : HGB(Histogram-based Gradient Boosting) in Titanic (Version 1.21)

Python : HGB 1.1 (2022.07.28)

  • HGB(Histogram-based Gradient Boosting) with some parameters' change
    I set max_iter=1000 in my dream last night …… omg 
    hgb = HistGradientBoostingClassifier(max_leaf_nodes=5, learning_rate=0.01, max_iter=1000, random_state=604)
  • Performance Scores (Accuracy)
    • Test (in train.csv) : 0.8435754189944135
    • Submission : 0.76555 (somewhat improved but I'm still thirsty!)
  • Kaggle Code : HGB(Histogram-based Gradient Boosting) in Titanic (Version 1.1)

Python : HGB 1.0 (2022.07.27)

  • HGB(Histogram-based Gradient Boosting) with default parameters
  • Use HistGradientBoostingClassifier() from sklearn
  • Pre-processing
    • Remove 4 variables : 1 PassengerId, 3 Name, 8 Ticket (useless) / 10 Cabin (too many NaN)
    • Replace 3 variables : 4 Sex(categorical) 5 Age(fill NaN) 11 Embarked(fill NaN, categorical)
  • Performance Scores (Accuracy)
    • Trainning : 0.9459309962075663
    • Validation : 0.8217275682064414
    • Test (in train.csv) : 0.8324022346368715
    • Submission : 0.75598 (disappointed ……)
  • Kaggle Code : HGB(Histogram-based Gradient Boosting) in Titanic (Version 1.0)