GitHub

SMOTENC

Initiate a SMOTENC model with the given hyper-parameters.

SMOTENC

A model type for constructing a smotenc, based on Imbalance.jl, and implementing the MLJ model interface.

From MLJ, the type can be imported using

SMOTENC = @load SMOTENC pkg=Imbalance

Do model = SMOTENC() to construct an instance with default hyper-parameters. Provide keyword arguments to override hyper-parameter defaults, as in SMOTENC(k=...).

SMOTENC implements the SMOTENC algorithm to correct for class imbalance as in N. V. Chawla, K. W. Bowyer, L. O.Hall, W. P. Kegelmeyer, “SMOTE: synthetic minority over-sampling technique,” Journal of artificial intelligence research, 321-357, 2002.

Training data

In MLJ or MLJBase, wrap the model in a machine by

mach = machine(model)

There is no need to provide any data here because the model is a static transformer.

Likewise, there is no need to fit!(mach).

For default values of the hyper-parameters, model can be constructed by

model = SMOTENC()

Hyperparameters

  • k=5: Number of nearest neighbors to consider in the SMOTENC algorithm. Should be within the range [1, n - 1], where n is the number of observations; otherwise set to the nearest of these two values.

  • ratios=1.0: A parameter that controls the amount of oversampling to be done for each class

    • Can be a float and in this case each non-majority class will be oversampled to the size of the majority class times the float; the majority class counts are preserved. By default, all classes are oversampled to the size of the majority class
    • Can be a dictionary mapping each class label to the float ratio for that class

  • knn_tree: Decides the tree used in KNN computations. Either "Brute" or "Ball". BallTree can be much faster but may lead to inaccurate results.

  • rng::Union{AbstractRNG, Integer}=default_rng(): Either an AbstractRNG object or an Integer seed to be used with Xoshiro if the Julia VERSION supports it. Otherwise, uses MersenneTwister`.

Transform Inputs

  • X: A table with element scitypes that subtype Union{Finite, Infinite}. Elements in nominal columns should subtype Finite (i.e., have scitype OrderedFactor or Multiclass) and elements in continuous columns should subtype Infinite (i.e., have scitype Count or Continuous).
  • y: An abstract vector of labels (e.g., strings) that correspond to the observations in X

Transform Outputs

  • Xover: A matrix or table that includes original data and the new observations due to oversampling. depending on whether the input X is a matrix or table respectively
  • yover: An abstract vector of labels corresponding to Xover

Operations

  • transform(mach, X, y): resample the data X and y using SMOTENC, returning both the new and original observations

Example

using MLJ
using ScientificTypes
import Imbalance

## set probability of each class
class_probs = [0.5, 0.2, 0.3]                         
num_rows = 100
num_continuous_feats = 3
## want two categorical features with three and two possible values respectively
num_vals_per_category = [3, 2]

## generate a table and categorical vector accordingly
X, y = Imbalance.generate_imbalanced_data(num_rows, num_continuous_feats; 
                                class_probs, num_vals_per_category, rng=42)                      
julia> Imbalance.checkbalance(y)
1: ▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 19 (39.6%) 
2: ▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 33 (68.8%) 
0: ▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 48 (100.0%) 

julia> ScientificTypes.schema(X).scitypes
(Continuous, Continuous, Continuous, Continuous, Continuous)
## coerce nominal columns to a finite scitype (multiclass or ordered factor)
X = coerce(X, :Column4=>Multiclass, :Column5=>Multiclass)

## load SMOTE-NC
SMOTENC = @load SMOTENC pkg=Imbalance

## wrap the model in a machine
oversampler = SMOTENC(k=5, ratios=Dict(0=>1.0, 1=> 0.9, 2=>0.8), rng=42)
mach = machine(oversampler)

## provide the data to transform (there is nothing to fit)
Xover, yover = transform(mach, X, y)

julia> Imbalance.checkbalance(yover)
2: ▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 38 (79.2%) 
1: ▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 43 (89.6%) 
0: ▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 48 (100.0%)