scripts/builtin/adasyn.dml

#-------------------------------------------------------------
#
# Licensed to the Apache Software Foundation (ASF) under one
# or more contributor license agreements.  See the NOTICE file
# distributed with this work for additional information
# regarding copyright ownership.  The ASF licenses this file
# to you under the Apache License, Version 2.0 (the
# "License"); you may not use this file except in compliance
# with the License.  You may obtain a copy of the License at
#
#   http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing,
# software distributed under the License is distributed on an
# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
# KIND, either express or implied.  See the License for the
# specific language governing permissions and limitations
# under the License.
#
#-------------------------------------------------------------

# Builtin function for handing class imbalance using Adaptive Synthetic Sampling (ADASYN)
# by Haibo He et. al. In International Joint Conference on Neural Networks (2008). 1322-1328
#
# INPUT:
# --------------------------------------------------------------------------------------
# X        Feature matrix [shape: n-by-m]
# Y        Class labels [shape: n-by-1]
# k        Number of nearest neighbors
# beta     Desired balance level after generation of synthetic data [0, 1]
# dth      Distribution threshold
# seed     Seed for randomized data point selection
# --------------------------------------------------------------------------------------
#
# OUTPUT:
# -------------------------------------------------------------------------------------
# Xp       Feature matrix of n original rows followed by G = (ml-ms)*beta synthetic rows
# Yp       Class labels aligned with output X
# -------------------------------------------------------------------------------------

m_adasyn = function(Matrix[Double] X, Matrix[Double] Y, Integer k = 2,
  Double beta = 1.0, Double dth = 0.9, Integer seed = -1)
  return (Matrix[Double] Xp, Matrix[Double] Yp)
{
  if(k < 1) {
    print("ADASYN: k should not be less than 1. Setting k value to default k = 1.")
    k = 1
  }

  # Preprocessing
  freq = t(table(Y, 1));
  minorIdx = as.scalar(rowIndexMin(freq))
  majorIdx = as.scalar(rowIndexMax(freq))

  # (Step 1)
  # Calculate the degree of class imbalance, where d in (0, 1]
  d = as.scalar(freq[1,minorIdx])/sum(freq)

  # (Step 2)
  # Check if imbalance is lower than predefined threshold
  print("ADASYN: class imbalance: " + d)

  if(d >= dth) {
      stop("ADASYN: Class imbalance not large enough.")
  }

  # (Step 2a)
  # Calculate number of synthetic data examples
  G = as.scalar(freq[1,majorIdx]-freq[1,minorIdx])*beta

  # (Step 2b)
  # For each x_i in non-majority class, find k nearest neighbors.
  # Get G random points from the KNN set via a permutation matrix multiply
  Xnonmajor = removeEmpty(target=X, margin="rows", select=(Y!=majorIdx))
  Ynonmajor = removeEmpty(target=Y, margin="rows", select=(Y!=majorIdx))
  NNR = knnbf(Xnonmajor, Xnonmajor, k+1)
  NNR = matrix(NNR, rows=length(NNR), cols=1)
  I = rand(rows=nrow(NNR), cols=1, seed=seed) < (G/nrow(NNR))
  NNRg = removeEmpty(target=NNR, margin="rows", select=I);
  P = table(seq(1, nrow(NNRg)), NNRg, nrow(NNRg), nrow(Xnonmajor));
  Xp = rbind(X, P %*% Xnonmajor);
  Yp = rbind(Y, P %*% Ynonmajor); # multi-class
}