/titanic-mogp

Multi Objective Genetic Programming Challenge on Titanic Dataset

Primary LanguageJupyter NotebookMIT LicenseMIT

Titanic MOGP

VIP Bootcamp

Working with Titanic dataset and multi objective GP. 🚢

Constraints

  1. We can:

    • Use strongly typed GP to predict boolean answers.
    • Use strongly typed GP to predict float answers, then convert the floats to booleans in your evaluation function (e.g., >0 or <0 for 0 and 1).
    • Work with loosely typed GP and manage the results in your evaluation function.

  2. Don't use methods from here.

    method name Usage/Returns
    deap.algorithms.eaSimple(population, toolbox, cxpb, mutpb, ngen[, stats, halloffame, verbose]) A class:~deap.tools.Logbook with the statistics of the evolution
    deap.algorithms.eaMuPlusLambda(population, toolbox, mu, lambda_, cxpb, mutpb, ngen[, stats, halloffame, verbose]) ^^
    deap.algorithms.eaMuCommaLambda(population, toolbox, mu, lambda_, cxpb, mutpb, ngen[, stats, halloffame, verbose]) ^^
    deap.algorithms.eaGenerateUpdate(toolbox, ngen[, stats, halloffame, verbose]) ^^
    deap.algorithms.varAnd(population, toolbox, cxpb, mutpb) A list of varied individuals that are independent of their parents.
    deap.algorithms.varOr(population, toolbox, lambda_, cxpb, mutpb) The final population.
    class deap.cma.Strategy(centroid, sigma[, **kargs])[source]
    class deap.cma.StrategyOnePlusLambda(parent, sigma[, **kargs])
    class deap.cma.StrategyMultiObjective(population, sigma[, **kargs])

  3. Upload a single CSV file with predictions on test.csv for each Pareto optimal individual you unearth from the train.csv in separate columns. Reuse the preprocessed and folded data from the Titanic ML assignment.

  4. On DEAP when we look at the fitness of our MOGA, we use individual.fitness.values to see how fit our individuals are. We must assign it in the eval function/main loop. By taking in the individual itself, we modify/return the mutant. We must build the truth data so that we can pass features through/get predictions.

📢 Presentation Details

  • Data Processing: EDA and our preprocessing from last assignment
  • ML Algorithms: On how we got the final Pareto-optimal set.
  • Evolutionary Algorithm Design: evolutionary algorithm's architecture. Describe algorithm design iterations, AAD focus
  • Comparison Time: BOBB on DEAP, we benchmark against our own models from last week, (random forest, xgboost, log regression, etc.)