saurabhmathur96/logistic-regression

An implementation of Logistic Regression in pure Java.

Logistic-Regression

An implementation of Binary Logistic Regression Classifier using Gradient Descent in pure Java 8.

Classes :

• Instance
• DataPreProcessor
• BinaryClassifier

Test-Data Files:

• test-scores-data.csv : From Stanford's ml-class on Coursera, assignment2 ex2data1.txt
• xor.csv : The classifier fails on this as the data is not linearly separable.

Output from test run:

Initial cost = 0.693147
== Starting Training ==
Learning Rate = 0.050000, No. of iterations = 40000
== Training Complete ==
BinaryClassifier [ biasWeight=0.8150521625489258 weights=[1.8105877596418987, 1.6405966520919257] ]
Final cost = 0.258550
Test data = [45.0, 85.0]
h(test) = 0.634277

Documentation

public class Instance

Stores a single training example for binary classification. Created by saurabh on 10/2/16.

public class DataPreProcessor

Created by saurabh on 12/2/16.

public DataPreProcessor(ArrayList<ArrayList<Double>> data)

computes the column-wise mean and standard deviation

• Parameters: data —

public ArrayList<Double> normalizeRow(ArrayList<Double> row)

Normalizes the row using mean and standard deviation. newX = x - mean/standardDeviation

• Parameters: row —
• Returns: normalizedRow (ArrayList)

public double[] normalizeRow(double[] row)

Normalizes the row using mean and standard deviation. newX = x - mean/standardDeviation

• Parameters: row —
• Returns: normalizedRow (array of doubles)

public ArrayList<Instance> createTrainingExamples(ArrayList<ArrayList<Double>> rows, ArrayList<Integer> labels)

Converts data to ArrayList of Instance(row, label) for training BinaryClassifier

• Parameters:
  • rows —
  • labels —
• Returns: trainingExamples (ArrayList of Instance objects)

public class BinaryClassifier

An implementation of logistic regression using gradient descent for 2 class classification. Created by saurabh on 10/2/16.

public double computeHypothesis(double[] inputs)

computes the probability that the given set of inputs have label 1.

• Parameters: inputs —
• Returns: probability that label = 1

public double computeCost(ArrayList<Instance> examples)

computes the cost function which is to be minimized.

• Parameters: examples —
• Returns: cost

public void train(ArrayList<Instance> examples, double rate, int nIterations, boolean verbose)

Optimises the weights using given training data with gradient descent. calls _trainIteration nIterations number of times.

• Parameters:
  • examples —
  • rate —
  • nIterations —
  • verbose —

public int classify(double[] inputs, double threshold)

classifies given inputs as 1 if hypothesis value >= threshold else 0

• Parameters:
  • inputs —
  • threshold —
• Returns: class 0 or 1