add basic functionality for training logistic regression

Author: aditsinha

.gitignore

@@ -0,0 +1 @@
+*~

main.cc

@@ -0,0 +1,25 @@
+
+#include <cassert>
+#include <iostream>
+#include <fstream>
+#include <string>
+#include <cstring>
+#include <cmath>
+
+#include <eigen3/Eigen/Dense>
+
+#include "party.h"
+#include "train.h"
+
+
+int main(int argc, char** argv) {
+  // the first argument is the configuration file
+  // the second argument is the data file
+
+  assert(argc == 3);
+
+  // For configuration file, first line contains n and m
+  std::string config_filename = argv[1];
+
+
+}

party.cc

@@ -0,0 +1,72 @@
+
+#include "party.h"
+
+Configuration::Configuration(std::ifstream config_file)  {
+  assert(config_file.is_open());
+
+  // First read the type of training we are going to do
+  config_file >> mode;
+  // Next read the size of the dataset (number of parties, number of
+  // entries per party, number of feature dimensions)
+  config_file >> n >> m >> d;
+  // For each feature dimension, read the normalization factor
+
+  normalization = Eigen::VectorXd(d);
+
+  for (int i = 0; i < d; i++) {
+    std::string str;
+    std::getline(config_file, str, ',');
+    normalization(i) = std::stof(str);
+  }
+}
+
+Party::Party(Configuration* config, std::ifstream data_file)
+  : features(config->n, config->m)
+{
+  assert(data_file.is_open());
+
+  n = config->n;
+  m = config->m;
+  d = config->d;
+
+  features = Eigen::MatrixXd(m, d);
+  labels = Eigen::VectorXd(m);
+  
+  // assume that data is in CSV format, where the first column is the
+  // label
+  
+  for (int j = 0; j < m; j++) {
+    std::string str;
+    // read the label
+    std::getline(data_file, str, ',');
+    labels(j) = std::stoi(str);
+
+    // now read the feature vector
+    for (int k = 0; k < d; k++) {
+      std::getline(data_file, str, ',');
+      features(j,k) = std::stof(str) * config->normalization(k);
+    }
+
+    // Check that the label and features are correct
+    assert(labels(j) == 0 || labels(j) == 1);
+    assert(getFeatureVec(j).norm() <= 1);
+  }
+}
+
+Eigen::VectorXd Party::getFeatureVec(int i) {
+  return features.block(i,0,1,d);
+}
+
+double ComputeLogisticFn(Eigen::VectorXd params, Eigen::VectorXd features) {
+  return 1 / (1 + exp(-1 * params.dot(features)));
+}
+
+Eigen::VectorXd Party::ComputeGradient(Eigen::VectorXd params) {
+  Eigen::VectorXd grad = Eigen::VectorXd::Zero(d);
+  for (int i = 0; i < m; i++) {
+    double error = ComputeLogisticFn(params, getFeatureVec(i)) - labels(i);
+    grad += error * getFeatureVec(i);
+  }
+
+  return grad;
+}

party.h

@@ -0,0 +1,34 @@
+
+#pragma once
+
+#include <cassert>
+#include <iostream>
+#include <fstream>
+#include <string>
+#include <cstring>
+#include <cmath>
+
+#include <eigen3/Eigen/Dense>
+
+class Configuration {
+public:
+  Configuration(std::ifstream config_file);
+  int n, m, d;
+  Eigen::MatrixXd normalization;
+  std::string mode;
+};
+
+class Party {
+public:
+  Party(Configuration* config, std::ifstream data_file);
+  ~Party();
+  float* ComputeGradient(float* params);
+
+  Eigen::MatrixXd features;
+  Eigen::MatrixXd labels;
+
+  Eigen::VectorXd ComputeGradient(Eigen::VectorXd params);
+  Eigen::VectorXd getFeatureVec(int i);
+  
+  int n, m, d;
+};

train.cc

@@ -0,0 +1,17 @@
+
+
+#include <eigen3/Eigen/Dense>
+
+#include "party.h"
+
+Eigen::VectorXd train_single(Party* p) {
+  Eigen::VectorXd params = Eigen::VectorXd::Zero(p.d);
+
+  learning_rate = .01;
+
+  for (int i = 0; i < 1000; i++) {
+    params = params - learning_rate * p->ComputeGradient(params);
+  }
+
+  return params;
+}

train.h

@@ -0,0 +1,7 @@
+
+#pragma once
+
+#include <eigen3/Eigen/Dense>
+#include "party.h"
+
+Eigen::VectorXd train_single(Party* p);