From b49d0793d4ea01a3a18bd596b9cde16c77bb7a97 Mon Sep 17 00:00:00 2001
From: Stephen Mayhew <mayhew2@illinois.edu>
Date: Mon, 4 Jan 2016 12:17:05 -0600
Subject: [PATCH] moved files to a new package, and updated readme
---
README.md | 7 +-
pom.xml | 36 +++++++++-
.../java/edu/illinois/cs/cogcomp/Tester.java | 27 --------
.../cs/cogcomp/{ => lm}/NeuralLM.java | 37 ++++++----
.../illinois/cs/cogcomp/{ => lm}/NplmJni.java | 2 +-
.../edu/illinois/cs/cogcomp/lm/Tester.java | 68 +++++++++++++++++++
6 files changed, 133 insertions(+), 44 deletions(-)
delete mode 100644 src/main/java/edu/illinois/cs/cogcomp/Tester.java
rename src/main/java/edu/illinois/cs/cogcomp/{ => lm}/NeuralLM.java (93%)
rename src/main/java/edu/illinois/cs/cogcomp/{ => lm}/NplmJni.java (87%)
create mode 100644 src/main/java/edu/illinois/cs/cogcomp/lm/Tester.java
diff --git a/README.md b/README.md
index 81bfa58..86eafd9 100644
--- a/README.md
+++ b/README.md
@@ -1,8 +1,13 @@
Neural Probabilistic Language Model Toolkit -- Java
===================================================
-This is based on the [NPLM toolkit](http://nlg.isi.edu/software/nplm/) from ISI. In fact, it is a simple port of the nplm.py python file from that distribution. The goal is to have a reader for the language model file which can be used in Java. The interesting part is the matrix multiplication to get the probability of text.
+This is based on the [NPLM toolkit](http://nlg.isi.edu/software/nplm/) from ISI. In fact, it
+is a simple port of the nplm.py python file from that distribution. The goal is to have a
+reader for the language model file which can be used in Java. The interesting part is the
+matrix multiplication to get the probability of text.
Hopefully there will be a JNI section for training a LM from Java also.
+To train a model, use all the tools available from the NPLM toolkit (not in Java). These are executables
+called `prepareNeuralNetwork`, `trainNeuralNetwork`, `testNeuralNetwork`, `testNeuralLM`, etc.
diff --git a/pom.xml b/pom.xml
index 925ce24..1e0ab8d 100644
--- a/pom.xml
+++ b/pom.xml
@@ -6,7 +6,7 @@
<groupId>edu.illinois.cs.cogcomp</groupId>
<artifactId>nplm-java</artifactId>
- <version>1.0-SNAPSHOT</version>
+ <version>1.0.0</version>
<repositories>
<repository>
@@ -63,4 +63,36 @@
</dependencies>
-</project>
\ No newline at end of file
+ <build>
+ <plugins>
+ <plugin>
+ <groupId>org.apache.maven.plugins</groupId>
+ <artifactId>maven-compiler-plugin</artifactId>
+ <version>2.0.2</version>
+ <configuration>
+ <source>1.7</source>
+ <target>1.7</target>
+ </configuration>
+ </plugin>
+ </plugins>
+
+ <extensions>
+ <extension>
+ <groupId>org.apache.maven.wagon</groupId>
+ <artifactId>wagon-ssh</artifactId>
+ <version>2.4</version>
+ </extension>
+ </extensions>
+
+ </build>
+
+ <distributionManagement>
+ <repository>
+ <id>CogcompSoftware</id>
+ <name>CogcompSoftware</name>
+ <url>scp://bilbo.cs.illinois.edu:/mounts/bilbo/disks/0/www/cogcomp/html/m2repo</url>
+ </repository>
+ </distributionManagement>
+ <packaging>jar</packaging>
+
+</project>
diff --git a/src/main/java/edu/illinois/cs/cogcomp/Tester.java b/src/main/java/edu/illinois/cs/cogcomp/Tester.java
deleted file mode 100644
index 7146d2e..0000000
--- a/src/main/java/edu/illinois/cs/cogcomp/Tester.java
+++ /dev/null
@@ -1,27 +0,0 @@
-package edu.illinois.cs.cogcomp;
-
-import org.apache.commons.math3.linear.*;
-
-import java.util.ArrayList;
-import java.util.List;
-
-/**
- * Created by mayhew2 on 11/19/15.
- */
-public class Tester {
-
- public static void main(String[] args) throws Exception {
-
- NeuralLM m = NeuralLM.from_file("/shared/experiments/mayhew2/transliteration/NEURAL_LANGUAGE_MODEL/example/inferno.nnlm");
-
- List<String> ngrams = new ArrayList<>();
-
- ngrams.add("fair");
- ngrams.add("and");
- ngrams.add("xkcd");
-
- System.out.println(m.ngram_prob(ngrams));
-
- }
-
-}
diff --git a/src/main/java/edu/illinois/cs/cogcomp/NeuralLM.java b/src/main/java/edu/illinois/cs/cogcomp/lm/NeuralLM.java
similarity index 93%
rename from src/main/java/edu/illinois/cs/cogcomp/NeuralLM.java
rename to src/main/java/edu/illinois/cs/cogcomp/lm/NeuralLM.java
index dcc001d..16bf989 100644
--- a/src/main/java/edu/illinois/cs/cogcomp/NeuralLM.java
+++ b/src/main/java/edu/illinois/cs/cogcomp/lm/NeuralLM.java
@@ -1,4 +1,4 @@
-package edu.illinois.cs.cogcomp;
+package edu.illinois.cs.cogcomp.lm;
import edu.illinois.cs.cogcomp.core.io.LineIO;
import org.apache.commons.math3.linear.*;
@@ -8,7 +8,6 @@ import org.apache.commons.math3.random.UniformRandomGenerator;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
-import java.io.FileNotFoundException;
import java.util.*;
/**
@@ -113,22 +112,34 @@ public class NeuralLM {
uniform(this.output_biases, r);
}
+
/**
- * inputs is a list of one-hot vectors, size of each should be input_embedding_dimension.
- * length of inputs is ngram-1. The first element of inputs is a matrix containing one-hot
- * vectors of the first word in each ngram. The second element is a matrix containing one-hot vectors
- * of the second word in each ngram. etc.
- *
- * output is a matrix containing one-hot vectors, where each vector is the last in each ngram.
- * output is also a one-hot vector, size of output_embedding_dimension.
- *
- * @return
+ * This is primarily intended for character based n-grams. Convenient.
+ * @param chars needs to have length of ngram_size.
+ * @return log probability of ngram.
+ * @throws Exception
+ */
+ public double ngram_prob(char[] chars) throws Exception {
+ List<String> lst = new ArrayList<>();
+ for(char c : chars){
+ lst.add(c + "");
+ }
+ return ngram_prob(lst);
+ }
+
+ /**
+ * get the probability of a single ngram
+ * @param ngramlist this is a list of items in the ngram. This needs to be of
+ * size ngram_size. For example, {'john', 'ate', 'cabbage'}
+ * for ngram_size=3
+ * @return the log probability of this ngram
+ * @throws Exception need to fix this.
*/
public double ngram_prob(List<String> ngramlist) throws Exception {
// the input here is a single ngram. Length of ngram
if(ngramlist.size() != ngram_size){
- throw new Exception("ngmram-list must have the same size as ngram_size");
+ throw new Exception("ngram-list must have the same size as ngram_size");
}
int unkInd = this.word_to_index.get("<unk>");
@@ -153,6 +164,7 @@ public class NeuralLM {
}
RealVector r = new ArrayRealVector(this.n_input_vocab);
r.setEntry(ind, 1);
+ // FIXME: this could be changed to a index, not a multiplication
RealVector r_embed = this.input_embeddings.transpose().operate(r);
concatenated = concatenated.append(r_embed);
}
@@ -347,7 +359,6 @@ public class NeuralLM {
}else if(section.equals("\\output_vocab")){
// FIXME: not sure there is anything I need to do???
} else if(section.equals("\\input_embeddings")) {
- logger.debug("got here");
m.input_embeddings = read_matrix(sectionlines, m.n_input_vocab, m.input_embedding_dimension);
}else if(section.equals("\\hidden_weights 1")) {
m.hidden1_weights = read_matrix(sectionlines, m.n_hidden, (m.ngram_size - 1) * m.input_embedding_dimension);
diff --git a/src/main/java/edu/illinois/cs/cogcomp/NplmJni.java b/src/main/java/edu/illinois/cs/cogcomp/lm/NplmJni.java
similarity index 87%
rename from src/main/java/edu/illinois/cs/cogcomp/NplmJni.java
rename to src/main/java/edu/illinois/cs/cogcomp/lm/NplmJni.java
index ef4d07c..0e8654b 100644
--- a/src/main/java/edu/illinois/cs/cogcomp/NplmJni.java
+++ b/src/main/java/edu/illinois/cs/cogcomp/lm/NplmJni.java
@@ -1,4 +1,4 @@
-package edu.illinois.cs.cogcomp;
+package edu.illinois.cs.cogcomp.lm;
/**
* Created by mayhew2 on 11/18/15.
diff --git a/src/main/java/edu/illinois/cs/cogcomp/lm/Tester.java b/src/main/java/edu/illinois/cs/cogcomp/lm/Tester.java
new file mode 100644
index 0000000..830671d
--- /dev/null
+++ b/src/main/java/edu/illinois/cs/cogcomp/lm/Tester.java
@@ -0,0 +1,68 @@
+package edu.illinois.cs.cogcomp.lm;
+
+import edu.illinois.cs.cogcomp.core.io.LineIO;
+import edu.illinois.cs.cogcomp.lm.NeuralLM;
+
+import java.util.ArrayList;
+import java.util.List;
+
+/**
+ * Created by mayhew2 on 11/19/15.
+ */
+public class Tester {
+
+ public static void main(String[] args) throws Exception {
+
+ NeuralLM m = NeuralLM.from_file("/home/mayhew2/IdeaProjects/illinois-transliteration/lm/nplm-ru.txt");
+
+// List<String> ngrams = new ArrayList<>();
+
+ //ngrams.add("Ñ‚");
+ //ngrams.add("и");
+ //ngrams.add("н");
+
+// ngrams.add("a");
+// ngrams.add("b");
+// ngrams.add("н");
+
+ List<String> lines = LineIO.read("/home/mayhew2/IdeaProjects/illinois-transliteration/lm/ruwords.txt");
+
+ double total = 0;
+
+ List<String> ngrams = new ArrayList<>();
+
+ for(String line : lines){
+ line = "<s> " + line;
+ String[] sline = line.split(" ");
+
+ for(int i = 0; i < sline.length-1; i++){
+ ngrams.clear();
+
+ ngrams.add(sline[i]);
+ ngrams.add(sline[i+1]);
+
+
+ total += m.ngram_prob(ngrams);
+ }
+
+
+ }
+ System.out.println(total);
+
+ System.out.println();
+ String russian = "хруÑталь";
+
+ for(char c : russian.toCharArray()){
+ ngrams.clear();
+ ngrams.add(c + "");
+ ngrams.add("</s>");
+
+ System.out.println(m.ngram_prob(ngrams));
+ }
+
+
+
+
+ }
+
+}
--
GitLab