apaam/mlpack
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Cleaned up and optimized naive_bayes a bit.

Browse Source 
Refactored code into separate header and implementation files.
Updated some documentation.
Removed some unecessary variables and uses less memory.
Should be slightly faster than trunk now.
This commit is contained in:
James Cline
2011-02-01 18:43:09 +00:00
parent 8d56cdfefe
commit 3a47821be9
7 changed files with 254 additions and 207 deletions
Download Patch File Download Diff File Expand all files Collapse all files
fastlib/branches/fastlib-stl/mlpack/naive_bayes/CMakeLists.txt
+2 -2
View File
@@ -4,8 +4,8 @@ cmake_minimum_required(VERSION 2.8)
# Anything not in this list will not be compiled into the output library
# Do not include test programs here
set(SOURCES
simple_nbc.h
phi.h
simple_nbc.cc
phi.cc
)
# add directory name to sources
fastlib/branches/fastlib-stl/mlpack/naive_bayes/README.txt
+4 -6
View File
@@ -6,15 +6,13 @@ The files are the following:
--train : the file that contains the training data, the last column being the class of the data point
--nbc/classes : the number of classes the data provided has been classified into
--test : this file contains the testing data, this still contains its actual labels on the last column, but it is not used.
--output : the file into which you want the output to be written into, defaults to "output.csv" (but it does not output in .csv format, it outputs in the "pretty-print" format)
--output : the file into which you want the output to be written into, defaults to "output.csv"
2. simple_nbc.h - this is the file that contains the definition of the class SimpleNaiveBayesClassifier. The rest of the details are present in the file itself.
3. phi.h - this contains the function that calculates the value of the univariate Gaussian pdf
4. math_functions.h - this file contains functions like returning the highest element and its index in an array/vector.
5. test_simple_nbc_main.cc - this file contains the class which tests the class SimpleNaiveBayesClassifier. The executable formed is "test_simple_nbc_main".
4. test_simple_nbc_main.cc - this file contains the class which tests the class SimpleNaiveBayesClassifier. The executable formed is "test_simple_nbc_main".
- the parameters taken in by this are the following:
--training_set : the training set, defaults to the file "trainSet.arff"
--training_results : the training results, defaults to the file "trainRes.arff"
@@ -22,9 +20,9 @@ The files are the following:
--testing_results : the testing results, defaults to the file "testRes.arff"
--num_classes : the number of classes in the data, defaults to "2", which in the number of classes of the data present in the default files
6. the .arff files, whose use has been described above.
5. the .arff files, whose use has been described above.
7. build.py - you might want to take a look at them for the purpose of seeing what all files are used where.
6. build.py - you might want to take a look at them for the purpose of seeing what all files are used where.
-> An example run would the following:
fl-build nbc_main
fastlib/branches/fastlib-stl/mlpack/naive_bayes/nbc_main.cc
+1 -3
View File
@@ -31,8 +31,6 @@
#include <fastlib/data/dataset.h>
#include <fastlib/base/arma_compat.h>
#include <iostream>
#include <fstream>
const fx_entry_doc parm_nbc_main_entries[] = {
{"train", FX_REQUIRED, FX_STR, NULL,
@@ -95,7 +93,7 @@ int main(int argc, char* argv[]) {
fx_timer_start(nbc_module, "testing");
////// Calling the function that classifies the test data
nbc.Classify(testing_data, &results);
nbc.Classify(testing_data, results);
fx_timer_stop(nbc_module, "testing");
fastlib/branches/fastlib-stl/mlpack/naive_bayes/phi.cc
+114
View File
@@ -0,0 +1,114 @@
/**
* @author Parikshit Ram (pram@cc.gatech.edu)
* @file phi.h
*
* This file computes the Gaussian probability
* density function
*/
#include "fastlib/fastlib.h"
#include "fastlib/fastlib_int.h"
#include <cmath>
#include "phi.h"
long double phi(const arma::vec& x, const arma::vec& mean, const arma::mat& cov) {
long double det, f;
double exponent;
index_t dim;
arma::mat inv;
arma::vec diff, tmp;
dim = x.n_rows;
inv = arma::inv(cov);
det = arma::det(cov);
if( det < 0){
det = -det;
}
diff = mean - x;
tmp = inv*diff;
exponent = arma::dot(diff,tmp);
long double tmp1, tmp2, tmp3;
tmp1 = 1;
tmp2 = dim;
tmp2 = tmp2/2;
tmp2 = pow((2*(math::PI)),tmp2);
tmp1 = tmp1/tmp2;
tmp3 = 1;
tmp2 = sqrt(det);
tmp3 = tmp3/tmp2;
tmp2 = -exponent;
tmp2 = tmp2 / 2;
f = (tmp1*tmp3*exp(tmp2));
return f;
}
long double phi(const double x, const double mean, const double var) {
long double f;
f = exp( -1.0*( (x-mean)*(x-mean)/(2*var) ) )/sqrt(2*math::PI*var);
return f;
}
long double phi(const arma::vec& x, const arma::vec& mean, const arma::mat& cov, const std::vector<arma::mat>& d_cov, arma::vec& g_mean, arma::vec& g_cov){
long double det, f;
double exponent;
index_t dim;
arma::mat inv;
arma::vec diff, tmp;
dim = x.n_rows;
inv = arma::inv(cov);
det = arma::det(cov);
if( det < 0){
det = -det;
}
diff = mean - x;
tmp = inv*diff;
exponent = arma::dot(diff,tmp);
long double tmp1, tmp2, tmp3;
tmp1 = 1;
tmp2 = dim;
tmp2 = tmp2/2;
tmp2 = pow((2*(math::PI)),tmp2);
tmp1 = tmp1/tmp2;
tmp3 = 1;
tmp2 = sqrt(det);
tmp3 = tmp3/tmp2;
tmp2 = -exponent;
tmp2 = tmp2 / 2;
f = (tmp1*tmp3*exp(tmp2));
// Calculating the g_mean values which would be a (1 X dim) vector
g_mean = f*tmp;
// Calculating the g_cov values which would be a (1 X (dim*(dim+1)/2)) vector
arma::vec g_cov_tmp(d_cov.size());
for(index_t i = 0; i < d_cov.size(); i++){
arma::vec tmp_d;
arma::mat inv_d;
long double tmp_d_cov_d_r;
tmp_d = d_cov[i]*tmp;
tmp_d_cov_d_r = arma::dot(tmp_d,tmp);
inv_d = inv*d_cov[i];
for(index_t j = 0; j < dim; j++)
tmp_d_cov_d_r += inv_d(j,j);
g_cov_tmp[i] = f*tmp_d_cov_d_r/2;
}
g_cov = g_cov_tmp;
return f;
}
fastlib/branches/fastlib-stl/mlpack/naive_bayes/phi.h
+8 -93
View File
@@ -5,6 +5,10 @@
* This file computes the Gaussian probability
* density function
*/
#ifndef MLPACK_PHI_H
#define MLPACK_PHI_H
#include "fastlib/fastlib.h"
#include "fastlib/fastlib_int.h"
#include <cmath>
@@ -21,39 +25,7 @@
* @endcode
*/
long double phi(arma::vec& x , arma::vec& mean , arma::mat& cov) {
long double det, f;
double exponent;
index_t dim;
arma::mat inv;
arma::vec diff, tmp;
dim = x.n_rows;
inv = arma::inv(cov);
det = arma::det(cov);
if( det < 0){
det = -det;
}
diff = mean - x;
tmp = inv*diff;
exponent = arma::dot(diff,tmp);
long double tmp1, tmp2, tmp3;
tmp1 = 1;
tmp2 = dim;
tmp2 = tmp2/2;
tmp2 = pow((2*(math::PI)),tmp2);
tmp1 = tmp1/tmp2;
tmp3 = 1;
tmp2 = sqrt(det);
tmp3 = tmp3/tmp2;
tmp2 = -exponent;
tmp2 = tmp2 / 2;
f = (tmp1*tmp3*exp(tmp2));
return f;
}
long double phi(const arma::vec& x, const arma::vec& mean, const arma::mat& cov);
/**
* Calculates the univariate Gaussian probability density function
@@ -66,13 +38,7 @@ long double phi(arma::vec& x , arma::vec& mean , arma::mat& cov) {
* @endcode
*/
long double phi(const double x, const double mean, const double var) {
long double f = 0.0;
f = exp(-1.0*((x-mean)*(x-mean)/(2*var)))/sqrt(2*math::PI*var);
return f;
}
long double phi(const double x, const double mean, const double var);
/**
* Calculates the multivariate Gaussian probability density function
@@ -87,57 +53,6 @@ long double phi(const double x, const double mean, const double var) {
* @endcode
*/
long double phi(arma::vec& x, arma::vec& mean, arma::mat& cov, std::vector<arma::mat>& d_cov, arma::vec *g_mean, arma::vec *g_cov){
long double det, f;
double exponent;
index_t dim;
arma::mat inv;
arma::vec diff, tmp;
dim = x.n_rows;
inv = arma::inv(cov);
inv = arma::inv(cov);
det = arma::det(cov);
long double phi(const arma::vec& x, const arma::vec& mean, const arma::mat& cov, const std::vector<arma::mat>& d_cov, arma::vec& g_mean, arma::vec& g_cov);
if( det < 0){
det = -det;
}
diff = mean - x;
tmp = inv*diff;
exponent = arma::dot(diff,tmp);
long double tmp1, tmp2, tmp3;
tmp1 = 1;
tmp2 = dim;
tmp2 = tmp2/2;
tmp2 = pow((2*(math::PI)),tmp2);
tmp1 = tmp1/tmp2;
tmp3 = 1;
tmp2 = sqrt(det);
tmp3 = tmp3/tmp2;
tmp2 = -exponent;
tmp2 = tmp2 / 2;
f = (tmp1*tmp3*exp(tmp2));
// Calculating the g_mean values which would be a (1 X dim) vector
*g_mean = f*tmp;
// Calculating the g_cov values which would be a (1 X (dim*(dim+1)/2)) vector
arma::vec g_cov_tmp(d_cov.size());
for(index_t i = 0; i < d_cov.size(); i++){
arma::vec tmp_d;
arma::mat inv_d;
long double tmp_d_cov_d_r;
tmp_d = d_cov[i]*tmp;
tmp_d_cov_d_r = arma::dot(tmp_d,tmp);
inv_d = inv*d_cov[i];
for(index_t j = 0; j < dim; j++)
tmp_d_cov_d_r += inv_d(j,j);
g_cov_tmp[i] = f*tmp_d_cov_d_r/2;
}
*g_cov = g_cov_tmp; // Is g_cov larger than g_cov_tmp??
// g_cov->Copy(g_cov_tmp,d_cov.size());
return f;
}
#endif // MLPACK_PHI_H
fastlib/branches/fastlib-stl/mlpack/naive_bayes/simple_nbc.cc
+122
View File
@@ -0,0 +1,122 @@
/**
* @author Parikshit Ram (pram@cc.gatech.edu)
* @file simple_nbc.h
*
* A Naive Bayes Classifier which parametrically
* estimates the distribution of the features.
* It is assumed that the features have been
* sampled from a Gaussian PDF
*
*/
#define ARMA_NO_DEBUG
#include "fastlib/fastlib.h"
#include "simple_nbc.h"
#include "phi.h"
void SimpleNaiveBayesClassifier::InitTrain(const arma::mat& data, datanode* nbc_module) {
index_t number_examples = data.n_cols;
index_t number_features = data.n_rows - 1;
nbc_module_ = nbc_module;
arma::vec feature_sum(number_features), feature_sum_squared(number_features);
// updating the variables, private and local, according to
// the number of features and classes present in the data
number_of_classes_ = fx_param_int_req(nbc_module_,"classes");
class_probabilities_.set_size(number_of_classes_);
means_.set_size(number_features,number_of_classes_);
variances_.set_size(number_features,number_of_classes_);
NOTIFY("%"LI"d examples with %"LI"d features each\n",
number_examples, number_features);
fx_result_int(nbc_module_, "features", number_features);
fx_result_int(nbc_module_, "examples", number_examples);
// calculating the class probabilities as well as the
// sample mean and variance for each of the features
// with respect to each of the labels
for(index_t i = 0; i < number_of_classes_; i++ ) {
index_t number_of_occurrences = 0;
for (index_t j = 0; j < number_examples; j++) {
index_t flag = (index_t) data(number_features, j);
if(i == flag) {
++number_of_occurrences;
for(index_t k = 0; k < number_features; k++) {
double tmp = data(k, j);
feature_sum(k) += tmp;
feature_sum_squared(k) += tmp*tmp;
}
}
}
class_probabilities_[i] = (double)number_of_occurrences
/ (double)number_examples ;
for(index_t k = 0; k < number_features; k++) {
double fs = feature_sum(k),
fss = feature_sum_squared(k);
means_(k, i) = (fs / number_of_occurrences);
variances_(k, i) = (fss
- (fs * fs / number_of_occurrences))
/(number_of_occurrences - 1);
/*
means_(k, i) = (feature_sum(k) / number_of_occurrences);
variances_(k, i) = (feature_sum_squared(k)
- (feature_sum(k) * feature_sum(k) / number_of_occurrences))
/(number_of_occurrences - 1);
*/
}
feature_sum.zeros(number_features);
feature_sum_squared.zeros(number_features);
}
}
void SimpleNaiveBayesClassifier::Classify(const arma::mat& test_data, arma::vec& results){
// Checking that the number of features in the test data is same
// as in the training data
DEBUG_ASSERT(test_data.n_rows - 1 == means_.n_rows);
arma::vec tmp_vals(number_of_classes_);
index_t number_features = test_data.n_rows - 1;
results.zeros(test_data.n_cols);
NOTIFY("%"LI"d test cases with %"LI"d features each\n",
test_data.n_cols, number_features);
fx_result_int(nbc_module_,"tests", test_data.n_cols);
// Calculating the joint probability for each of the data points
// for each of the classes
// looping over every test case
for (index_t n = 0; n < test_data.n_cols; n++) {
//looping over every class
for (index_t i = 0; i < number_of_classes_; i++) {
// Using the log values to prevent floating point underflow
tmp_vals(i) = log(class_probabilities_(i));
//looping over every feature
for (index_t j = 0; j < number_features; j++) {
tmp_vals(i) += log(phi(test_data(j, n),
means_(j, i),
variances_(j, i))
);
}
}
// Find the index of the maximum value in tmp_vals.
index_t max = 0;
for (index_t k = 0; k < number_of_classes_; k++) {
if(tmp_vals(max) < tmp_vals(k))
max = k;
}
results(n) = max;
}
return;
}
fastlib/branches/fastlib-stl/mlpack/naive_bayes/simple_nbc.h
+3 -103
View File
@@ -83,7 +83,7 @@ class SimpleNaiveBayesClassifier {
arma::mat means_, variances_;
// The variable containing the class probabilities
std::vector<double> class_probabilities_;
arma::vec class_probabilities_;
// The variable keeping the information about the
// number of classes present
@@ -94,11 +94,6 @@ class SimpleNaiveBayesClassifier {
public:
SimpleNaiveBayesClassifier(){
/*
means_.Init(0, 0);
variances_.Init(0, 0);
class_probabilities_.Init(0);
*/
}
~SimpleNaiveBayesClassifier(){
@@ -116,58 +111,7 @@ class SimpleNaiveBayesClassifier {
* nbc.InitTrain(training_data, nbc_module);
* @endcode
*/
void InitTrain(const arma::mat& data, datanode* nbc_module) {
std::vector<double> feature_sum, feature_sum_squared;
index_t number_examples = data.n_cols;
index_t number_features = data.n_rows - 1;
nbc_module_ = nbc_module;
// updating the variables, private and local, according to
// the number of features and classes present in the data
number_of_classes_ = fx_param_int_req(nbc_module_,"classes");
class_probabilities_.resize(number_of_classes_,0);
means_.zeros(number_features,number_of_classes_);
variances_.zeros(number_features,number_of_classes_);
feature_sum.resize(number_features,0);
feature_sum_squared.resize(number_features,0);
for(index_t k = 0; k < number_features; k++) {
feature_sum[k] = 0;
feature_sum_squared[k] = 0;
}
NOTIFY("%"LI"d examples with %"LI"d features each\n",
number_examples, number_features);
fx_result_int(nbc_module_, "features", number_features);
fx_result_int(nbc_module_, "examples", number_examples);
// calculating the class probabilities as well as the
// sample mean and variance for each of the features
// with respect to each of the labels
for(index_t i = 0; i < number_of_classes_; i++ ) {
index_t number_of_occurrences = 0;
for (index_t j = 0; j < number_examples; j++) {
index_t flag = (index_t) data(number_features, j);
if(i == flag) {
++number_of_occurrences;
for(index_t k = 0; k < number_features; k++) {
double tmp = data(k, j);
feature_sum[k] += tmp;
feature_sum_squared[k] += tmp*tmp;
}
}
}
class_probabilities_[i] = (double)number_of_occurrences
/ (double)number_examples ;
for(index_t k = 0; k < number_features; k++) {
means_(k, i) = (feature_sum[k] / number_of_occurrences);
variances_(k, i) = (feature_sum_squared[k]
- (feature_sum[k] * feature_sum[k] / number_of_occurrences))
/(number_of_occurrences - 1);
feature_sum[k] = 0;
feature_sum_squared[k] = 0;
}
}
}
void InitTrain(const arma::mat& data, datanode* nbc_module);
/**
* Given a bunch of data points, this function evaluates the class
@@ -180,50 +124,6 @@ class SimpleNaiveBayesClassifier {
* nbc.Classify(test_data, &results);
* @endcode
*/
void Classify(const arma::mat& test_data, arma::vec *results){
// Checking that the number of features in the test data is same
// as in the training data
DEBUG_ASSERT(test_data.n_rows - 1 == means_.n_rows);
arma::vec tmp_vals(number_of_classes_);
index_t number_features = test_data.n_rows - 1;
arma::vec evaluated_result(test_data.n_cols);
NOTIFY("%"LI"d test cases with %"LI"d features each\n",
test_data.n_cols, number_features);
fx_result_int(nbc_module_,"tests", test_data.n_cols);
// Calculating the joint probability for each of the data points
// for each of the classes
// looping over every test case
for (index_t n = 0; n < test_data.n_cols; n++) {
//looping over every class
for (index_t i = 0; i < number_of_classes_; i++) {
// Using the log values to prevent floating point underflow
tmp_vals[i] = log(class_probabilities_[i]);
for (index_t j = 0; j < number_features; j++) {
tmp_vals[i] += log(phi(test_data(j, n),
means_(j, i),
variances_(j, i))
);
}
}
// Find the index of the maximum value in tmp_vals.
evaluated_result[n] = 0;
for (index_t k = 0; k < number_of_classes_; k++) {
if(tmp_vals[evaluated_result[n]] < tmp_vals[k])
evaluated_result[n] = k;
}
}
// The result is being put in a vector
*results = evaluated_result;
return;
}
void Classify(const arma::mat& test_data, arma::vec& results);
};
#endif