public static LDASampler FastLoad(string lda_sampler_filename, int[][] WORDS_IN_DOCS)
{
using (FileStream fs = new FileStream(lda_sampler_filename, FileMode.Open))
{
using (BinaryReader br = new BinaryReader(fs))
{
int NUM_TOPICS = br.ReadInt32();
int NUM_WORDS = br.ReadInt32();
int NUM_DOCS = br.ReadInt32();
double ALPHA = br.ReadDouble();
double BETA = br.ReadDouble();
LDASampler lda_sampler = new LDASampler(ALPHA, BETA, NUM_TOPICS, NUM_WORDS, NUM_DOCS, WORDS_IN_DOCS, false);
lda_sampler.total_iterations = br.ReadInt32();
//internal int[][] topic_of_word_in_doc; // [doc][i]
for (int doc = 0; doc < NUM_DOCS; ++doc)
{
for (int i = 0; i < lda_sampler.topic_of_word_in_doc[doc].Length; ++i)
{
lda_sampler.topic_of_word_in_doc[doc][i] = br.ReadInt32();
}
}
//internal float[,] number_of_times_doc_has_a_specific_topic; // [doc,topic]
for (int doc = 0; doc < NUM_DOCS; ++doc)
{
for (int topic = 0; topic < NUM_TOPICS; ++topic)
{
lda_sampler.number_of_times_doc_has_a_specific_topic[doc, topic] = br.ReadSingle();
}
}
//internal float[] number_of_times_a_doc_has_any_topic; // [doc]
for (int doc = 0; doc < NUM_DOCS; ++doc)
{
lda_sampler.number_of_times_a_doc_has_any_topic[doc] = br.ReadSingle();
}
//internal float[,] number_of_times_a_topic_has_a_specific_word; // [topic,word]
for (int topic = 0; topic < NUM_TOPICS; ++topic)
{
for (int word = 0; word < NUM_WORDS; ++word)
{
lda_sampler.number_of_times_a_topic_has_a_specific_word[topic, word] = br.ReadSingle();
}
}
//internal float[] number_of_times_a_topic_has_any_word; // [topic]
for (int topic = 0; topic < NUM_TOPICS; ++topic)
{
lda_sampler.number_of_times_a_topic_has_any_word[topic] = br.ReadSingle();
}
return(lda_sampler);
}
}
}
public LDASamplerMCSerial(LDASampler lda_sampler, int NUM_THREADS)
{
this.lda_sampler = lda_sampler;
this.NUM_THREADS = NUM_THREADS;
// Work out the ratio of each doc for our sampling
{
int MAX_REPRESENTATION = 100;
int max_doc_length = 0;
for (int doc = 0; doc < lda_sampler.NUM_DOCS; ++doc)
{
max_doc_length = Math.Max(max_doc_length, lda_sampler.WORDS_IN_DOCS[doc].Length);
}
total_words_in_corpus = 0;
for (int doc = 0; doc < lda_sampler.NUM_DOCS; ++doc)
{
total_words_in_corpus += lda_sampler.WORDS_IN_DOCS[doc].Length;
}
random_mc_orderings = new List <int>();
for (int doc = 0; doc < lda_sampler.NUM_DOCS; ++doc)
{
int doc_representation = (0 < max_doc_length) ? MAX_REPRESENTATION * lda_sampler.WORDS_IN_DOCS[doc].Length / max_doc_length : 1;
if (0 == doc_representation && lda_sampler.WORDS_IN_DOCS[doc].Length > 0)
{
//Logging.Info("We have had to bump up the representation for doc {0} because it is too small", doc);
doc_representation = 1;
}
for (int i = 0; i < doc_representation; ++i)
{
random_mc_orderings.Add(doc);
}
}
}
random_mt = new RandomAugmented[NUM_THREADS];
probability_working_buffer = new double[NUM_THREADS][];
for (int thread = 0; thread < NUM_THREADS; ++thread)
{
random_mt[thread] = new RandomAugmented((DateTime.UtcNow.Millisecond * (1 + thread)));
probability_working_buffer[thread] = new double[lda_sampler.NUM_TOPICS];
}
}
public static void TestFastSaveLoad()
{
// Random data
double alpha = 0.1;
double beta = 0.2;
int num_topics = 200;
int num_words = 10000;
int num_docs = 20000;
int[][] WORDS_IN_DOCS = new int[num_docs][];
for (int i = 0; i < num_docs; ++i)
{
WORDS_IN_DOCS[i] = new int[i % 100 + 1];
}
LDASampler lda_sampler = new LDASampler(alpha, beta, num_topics, num_words, num_docs, WORDS_IN_DOCS);
{
string lda_sampler_filename = @"C:\temp\ldatest_old.dat";
Logging.Info("+OldSave");
SerializeFile.Save(lda_sampler_filename, lda_sampler);
Logging.Info("-OldSave");
Logging.Info("+OldLoad");
lda_sampler = (LDASampler)SerializeFile.Load(lda_sampler_filename);
Logging.Info("-OldLoad");
}
{
string lda_sampler_filename = @"C:\temp\ldatest_new.dat";
Logging.Info("+NewSave");
lda_sampler.FastSave(lda_sampler_filename);
Logging.Info("-NewSave");
Logging.Info("+NewLoad");
lda_sampler = FastLoad(lda_sampler_filename, WORDS_IN_DOCS);
Logging.Info("-NewLoad");
}
}
public LDAAnalysis(LDASampler lda)
{
this.lda = lda;
}
