/**
* init parameter for continue estimating or for later inference
*/
public bool initEstimatedModel(LDACommandLineOptions option)
{
if (!init(option))
{
return(false);
}
int m, n, w;
p = new double[K];
// load model, i.e., read z and trndata
if (!loadModel())
{
Console.WriteLine("Fail to load word-topic assignment file of the model!\n");
return(false);
}
Console.WriteLine("Model loaded:");
Console.WriteLine("\talpha:" + alpha);
Console.WriteLine("\tbeta:" + beta);
Console.WriteLine("\tM:" + M);
Console.WriteLine("\tV:" + V);
nw = ArrayInitializers.ZerosInt(V, K);
nd = ArrayInitializers.ZerosInt(M, K);
nwsum = ArrayInitializers.ZerosInt(K);
ndsum = ArrayInitializers.ZerosInt(M);
for (m = 0; m < data.M; m++)
{
int N = data.Docs[m].Length;
// assign values for nw, nd, nwsum, and ndsum
for (n = 0; n < N; n++)
{
w = data.Docs[m].Words[n];
int topic = z[m][n];
// number of instances of word i assigned to topic j
nw[w][topic] += 1;
// number of words in document i assigned to topic j
nd[m][topic] += 1;
// total number of words assigned to topic j
nwsum[topic] += 1;
}
// total number of words in document i
ndsum[m] = N;
}
theta = ArrayInitializers.Empty(M, K);
phi = ArrayInitializers.Empty(K, V);
dir = option.dir;
savestep = option.savestep;
return(true);
}
public bool init(LDACommandLineOptions option)
{
this.option = option;
trnModel = new Model();
trnModel.dfile = option.dfile;
trnModel.dir = option.dir;
trnModel.K = option.K;
trnModel.savestep = option.savestep;
trnModel.niters = option.niters;
if (option.est)
{
if (!trnModel.initNewModel(option))
{
return(false);
}
trnModel.data.LocalDictionary.WriteWordMap(option.dir + "\\" + option.wordMapFileName);
}
else if (option.estc)
{
if (!trnModel.initEstimatedModel(option))
{
return(false);
}
}
return(true);
}
//---------------------------------------------------------------
// Init Methods
//---------------------------------------------------------------
/**
* initialize the model
*/
protected bool init(LDACommandLineOptions option)
{
if (option == null)
{
return(false);
}
modelName = option.modelName;
K = option.K;
alpha = option.alpha;
if (alpha < 0.0)
{
alpha = 50.0 / K;
}
if (option.beta >= 0)
{
beta = option.beta;
}
niters = option.niters;
dir = option.dir;
if (dir.EndsWith("\\"))
{
dir = dir.Substring(0, dir.Length - 1);
}
dfile = option.dfile;
twords = option.twords;
wordMapFile = option.wordMapFileName;
return(true);
}
//-----------------------------------------------------
// Init method
//-----------------------------------------------------
public bool init(LDACommandLineOptions option)
{
this.option = option;
trnModel = new Model();
if (!trnModel.initEstimatedModel(option))
{
return(false);
}
globalDict = trnModel.data.LocalDictionary;
computeTrnTheta();
computeTrnPhi();
return(true);
}
/**
* Init parameters for inference
* reading new dataset from file
*/
public bool initNewModel(LDACommandLineOptions option, Model trnModel)
{
if (!init(option))
{
return(false);
}
LDADataset dataset = LDADataset.ReadDataset(dir + "\\" + dfile, trnModel.data.LocalDictionary);
if (dataset == null)
{
Console.WriteLine("Fail to read dataset!\n");
return(false);
}
return(initNewModel(option, dataset, trnModel));
}
public static void showHelp(LDACommandLineOptions option)
{
Console.WriteLine("LDA [options ...] [arguments...] \n");
Console.WriteLine(option.GetUsage());
}
static void Main(string[] args)
{
LDACommandLineOptions option = new LDACommandLineOptions();
var parser = new Parser();
option.beta = 0.1;
option.K = 10;
option.niters = 1000;
option.savestep = 100;
option.twords = 20;
option.dfile = "trndocs.dat";
option.dir = @"C:\Users\Amine\Documents\visual studio 2013\Projects\GibbsLDA.NET\GibbsLDA.NET\data";
option.est = true;
option.modelName = "model-final";
option.wordMapFileName = "wordmap.txt";
var stopWatch = new Stopwatch();
stopWatch.Start();
try
{
// if (args.length == 0){
// showHelp(parser);
// return;
// }
parser.ParseArguments(args, option);
if (option.est || option.estc)
{
Estimator estimator = new Estimator();
estimator.init(option);
estimator.estimate();
}
else if (option.inf)
{
Inferencer inferencer = new Inferencer();
inferencer.init(option);
Model newModel = inferencer.inference();
for (int i = 0; i < newModel.phi.Length; ++i)
{
//phi: K * V
Console.WriteLine("-----------------------\ntopic" + i + " : ");
for (int j = 0; j < 10; ++j)
{
Console.WriteLine(inferencer.globalDict.Id2Word[j] + "\t" + newModel.phi[i][j]);
}
}
}
}
catch (ParserException cle)
{
Console.WriteLine("Command line error: " + cle.Message);
showHelp(option);
Console.ReadLine();
return;
}
catch (Exception e)
{
Console.WriteLine("Error in main: " + e.Message);
Console.WriteLine(e.StackTrace);
Console.ReadLine();
return;
}
stopWatch.Stop();
Console.WriteLine("\n This run took : " + stopWatch.ElapsedMilliseconds / 1000.0 + " seconds");
Console.ReadLine();
}
/**
* Init parameters for inference
* @param newData DataSet for which we do inference
*/
public bool initNewModel(LDACommandLineOptions option, LDADataset newData, Model trnModel)
{
if (!init(option))
{
return(false);
}
int m, n;
var rnd = new Random();
K = trnModel.K;
alpha = trnModel.alpha;
beta = trnModel.beta;
p = new double[K];
Console.WriteLine("K:" + K);
data = newData;
//+ allocate memory and assign values for variables
M = data.M;
V = data.V;
dir = option.dir;
savestep = option.savestep;
Console.WriteLine("M:" + M);
Console.WriteLine("V:" + V);
// K: from command line or default value
// alpha, beta: from command line or default values
// niters, savestep: from command line or default values
nw = ArrayInitializers.ZerosInt(V, K);
nd = ArrayInitializers.ZerosInt(M, K);
nwsum = ArrayInitializers.ZerosInt(K);
ndsum = ArrayInitializers.ZerosInt(M);
z = new List <int> [M];
for (m = 0; m < data.M; m++)
{
int N = data.Docs[m].Length;
z[m] = new List <int>();
//initilize for z
for (n = 0; n < N; n++)
{
int topic = (int)Math.Floor(rnd.NextDouble() * K);
z[m].Add(topic);
// number of instances of word assigned to topic j
nw[data.Docs[m].Words[n]][topic] += 1;
// number of words in document i assigned to topic j
nd[m][topic] += 1;
// total number of words assigned to topic j
nwsum[topic] += 1;
}
// total number of words in document i
ndsum[m] = N;
}
theta = ArrayInitializers.Empty(M, K);
phi = ArrayInitializers.Empty(K, V);
return(true);
}
/**
* Init parameters for estimation
*/
public bool initNewModel(LDACommandLineOptions option)
{
//if (!init(option))
//return false;
var rnd = new Random();
int m, n, w, k;
p = new double[K];
data = LDADataset.ReadDataset(dir + "\\" + dfile);
if (data == null)
{
Console.WriteLine("Fail to read training data!\n");
return(false);
}
//+ allocate memory and assign values for variables
M = data.M;
V = data.V;
dir = option.dir;
savestep = option.savestep;
// K: from command line or default value
// alpha, beta: from command line or default values
// niters, savestep: from command line or default values
nw = new int[V][];
for (w = 0; w < V; w++)
{
nw[w] = new int[K];
for (k = 0; k < K; k++)
{
nw[w][k] = 0;
}
}
nd = new int[M][];
for (m = 0; m < M; m++)
{
nd[m] = new int[K];
for (k = 0; k < K; k++)
{
nd[m][k] = 0;
}
}
nwsum = new int[K];
for (k = 0; k < K; k++)
{
nwsum[k] = 0;
}
ndsum = new int[M];
for (m = 0; m < M; m++)
{
ndsum[m] = 0;
}
z = new List <int> [M];
for (m = 0; m < data.M; m++)
{
int N = data.Docs[m].Length;
z[m] = new List <int>();
//initilize for z
for (n = 0; n < N; n++)
{
int topic = (int)Math.Floor(rnd.NextDouble() * K);
z[m].Add(topic);
// number of instances of word assigned to topic j
nw[data.Docs[m].Words[n]][topic] += 1;
// number of words in document i assigned to topic j
nd[m][topic] += 1;
// total number of words assigned to topic j
nwsum[topic] += 1;
}
// total number of words in document i
ndsum[m] = N;
}
theta = new double[M][];
for (m = 0; m < M; m++)
{
theta[m] = new double[K];
}
phi = new double[K][];
for (k = 0; k < K; k++)
{
phi[k] = new double[V];
}
return(true);
}
