ModelMiner.jl

One package to train them all.

ModelMiner.jl aims to provide an easy use interface to train multiple machine learning models with a single function all. It provides a mine(...) interface through which users can train multiple classification and regression models easily. The goal is to quickly verify which models performs better on data without much effort.

Usage

Classification example

1. Prepare your data

using MLJ, RDatasets
data = dataset("datasets", "iris")
y, X = MLJ.unpack(data, ==(:Species))

2. Call mine(...) to train and evaluate models

using ModelMiner
results = mine(X, y)

The output will a dataframe with performance measures according to the data

7×3 DataFrame
 Row │ name                             Accuracy  MulticlassFScore 
     │ String                           Float64   Float64          
─────┼─────────────────────────────────────────────────────────────
   1 │ AdaBoostStumpClassifier          0.94              0.927209
   2 │ ConstantClassifier               0.266667          0.140233
   3 │ DecisionTreeClassifier           0.966667          0.965303
   4 │ DecisionTreeClassifier           0.92              0.918642
   5 │ DeterministicConstantClassifier  0.266667          0.140063
   6 │ KernelPerceptron                 0.94              0.937882
   7 │ LinearPerceptron                 0.813333          0.771626
   8 │ LinearSVC                        0.96              0.963594
   9 │ LogisticClassifier               0.973333          0.968547
  10 │ MultinomialClassifier            0.953333          0.952808
  11 │ NeuralNetworkClassifier          0.533333          0.405575
  12 │ NuSVC                            0.966667          0.96627
  13 │ Pegasos                          0.6               0.484509
  14 │ RandomForestClassifier           0.953333          0.957148
  15 │ RandomForestClassifier           0.953333          0.948225
  16 │ SVC                              0.96              0.957057
  17 │ XGBoostClassifier                0.94              0.939973

Regression example

1. Prepare your data

data = dataset("MASS", "Boston")
y, X = MLJ.unpack(data, ==(:MedV))

2. Train models and evaluate models

using ModelMiner
results = mine(X, y)

Results:

8×2 DataFrame
 Row │ name                            RootMeanSquaredError 
     │ String                          Float64              
─────┼──────────────────────────────────────────────────────
   1 │ ConstantRegressor                            9.22236
   2 │ DecisionTreeRegressor                        5.0619
   3 │ DecisionTreeRegressor                        4.51225
   4 │ DeterministicConstantRegressor               9.19367
   5 │ GaussianMixtureRegressor                     8.18446
   6 │ NeuralNetworkRegressor                      20.2011
   7 │ RandomForestRegressor                        3.98226
   8 │ RandomForestRegressor                        3.38003

TODO

• Automated Feature scaling
• Hyperparameter Tuning
• Reporting performance statistics from Cross Validations

Acknowledgements

Thanks to the developers of LazyPredict for inspiration and @ablaom for helping me overcome some challenges in developing.