MLPB is meant to become an organized collection of machine learning problems and solutions. In practice, machine learning often goes like this
I have this problem... I need to classify something as A, B or C using a combination of numeric and categorical features. If I could find a similar problem, maybe I could modify the solution to work for my needs.
This is where MLPB steps in. Want to see ML problems with sparse data? Got it. Want to compare Scikit-learn’s RandomForestRegressor with R’s randomForest? No problem. Need an example of predicting a ranked target variable? You’ve come to the right place.
MLPB contains a directory of Projects. Within each project is a designated Data directory and one or more machine learning problem directories. This looks something like
Projects/
Iris/
Data/
iris.csv
PredictSpecies/
predict_species_rf.R
predict_species_rf.py
predict_species_xgb.R
PredictSepalLength/
iris_sepal_length_xgb.R
JobTitles/
Data/
jobtitles.csv
PredictJobCategory/
naive_bayes_model.py
naive_bayes_model.R
Most of these directories should include a README.md file providing details about the data and/or problem. A single dataset can have have multiple problems, and a single problem can have multiple example models/solutions.
You can browse all the datasets and problems in MLPB's wiki. There, you can also search for problems with specific tags like [mult-class classification], [sparse-data], [NLP], etc.