See the implementation instructions below.
Note that the ui components rely on javafx, so java 8 is required.
You can download the Jar file directly from the latest release. Alternatively, you can build it yourself.
Just add the jar as external library. Be sure to copy the neuralnetwork.properties file to your resources folder.
Add following snippets to your pom.xml file to import the library:
<properties>
<!-- base properties -->
<maven.compiler.source>1.8</maven.compiler.source>
<maven.compiler.target>1.8</maven.compiler.target>
<project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>
<!-- paths -->
<main.class>your.main.class</main.class> <!-- change this line -->
<outputDir>target/classes</outputDir>
<resourceDir>src/main/resources</resourceDir>
<!-- plugin handling -->
<maven-dependency-plugin.version>3.2.0</maven-dependency-plugin.version>
<maven-assembly-plugin.version>3.1.0</maven-assembly-plugin.version>
<neuralnetwork.group>ch.kaiki.nn</neuralnetwork.group>
<neuralnetwork.id>neural-network</neuralnetwork.id>
<neuralnetwork.version>3.0</neuralnetwork.version> <!-- fixed version -->
<!-- <neuralnetwork.version>LATEST</neuralnetwork.version> --> <!-- latest version -->
</properties>
<repositories>
<!-- download url to sync with online repository -->
<repository>
<id>neuralnetwork</id>
<url>https://github.com/lpapailiou/neuralnetwork/raw/master</url>
</repository>
</repositories>
<dependencies>
<!-- maven dependency connecting to .m2 local repository -->
<dependency>
<groupId>${neuralnetwork.group}</groupId>
<artifactId>${neuralnetwork.id}</artifactId>
<version>${neuralnetwork.version}</version>
</dependency>
</dependencies>
Please note the neuralnetwork.properties file should be imported as well to your resources folder.
In case it is not present, you may include following plugin to your pom.xml file:
<build>
<outputDirectory>${outputDir}</outputDirectory>
<resources><resource><directory>${resourceDir}</directory></resource></resources>
<plugins>
<!-- copies properties file from neural network library to resources folder -->
<plugin>
<groupId>org.apache.maven.plugins</groupId>
<artifactId>maven-dependency-plugin</artifactId>
<version>${maven-dependency-plugin.version}</version>
<executions>
<execution>
<id>resource-dependencies</id>
<phase>compile</phase>
<goals>
<goal>unpack</goal>
</goals>
<configuration>
<artifactItems>
<artifactItem>
<groupId>${neuralnetwork.group}</groupId>
<artifactId>${neuralnetwork.id}</artifactId>
<version>${neuralnetwork.version}</version>
<type>jar</type>
<overWrite>true</overWrite>
</artifactItem>
</artifactItems>
<includes>**/*.properties</includes>
<outputDirectory>${project.basedir}/${resourceDir}</outputDirectory>
</configuration>
</execution>
</executions>
</plugin>
<!-- creates executable jar file with dependencies -->
<plugin>
<groupId>org.apache.maven.plugins</groupId>
<artifactId>maven-assembly-plugin</artifactId>
<version>${maven-assembly-plugin.version}</version>
<configuration>
<archive>
<manifest>
<mainClass>${main.class}</mainClass>
</manifest>
</archive>
<descriptorRefs>
<descriptorRef>jar-with-dependencies</descriptorRef>
</descriptorRefs>
</configuration>
<executions>
<execution>
<id>make-assembly</id>
<phase>package</phase>
<goals>
<goal>single</goal>
</goals>
</execution>
</executions>
</plugin>
</plugins>
</build>
and run mvn compile to complete the import of the properties file.
Below an overview of the neuralnetwork.properties file.
# ************************************************************************************************************ #
# *********** COMMON PROPERTIES *********** #
# ************************************************************************************************************ #
# the initializer function to initialize neural network matrices.
# available values: static|random|xavier|kaiming
initializer=random
# the learning rate must have a value between 0.0 and 1.0.
learning_rate=0.8
# the default rectifier as activation function for the neural network.
# available values: identity|relu|leaky_relu|sigmoid|sigmoid_accurate|silu|silu_accurate|tanh|elu|gelu|softplus|softmax.
rectifier=sigmoid
# the optimizer for the learning rate between iterations.
# available values: none|sgd
learning_rate_optimizer=sgd
# the learning rate decay as momentum.
# if learning_rate_optimizer is set to 'none', this value will have no effect.
# must have a value between 0.0 and 1.0.
learning_rate_momentum=0.01
# ************************************************************************************************************ #
# *********** SUPERVISED LEARNING ONLY *********** #
# ************************************************************************************************************ #
# the cost function is the metric for the error of one iteration. Its derivation is the loss.
# available values: mse_naive|mse|cross_entropy|exponential|hellinger_distance|kld|gkld|isd
cost_function=mse
# the regularizer will adapt the penalty by the loss function.
# available values: none|l1|l2|elastic
regularizer=none
# regularization functions may rely on a regularization parameter. this parameter, usually called lambda,
# will be controlled by the property regularizer_param.
regularizer_param=0
# dropout is an additional regularization technique. during training, it will set a certain percentage
# of output layer weights to zero and scale the output values by that factor.
# during testing, dropout is not active, the output will be scaled up instead.
dropout_factor=0
# batch mode decides if gradients are summed up or the mean is used for backpropagation.
# available values: mean|sum
batch_mode=mean
# ************************************************************************************************************ #
# *********** GENETIC ALGORITHM ONLY *********** #
# ************************************************************************************************************ #
# the parent count is the count of NeuralNetworks chosen for reproduction to be seeded to
# the new generation to come.
genetic_parent_count=1
# the selection pool size for genetic evolution as a percentage of the best performing neural networks. fallback will be 1.
genetic_reproduction_pool_size=0.5
# the mutation rate is the percentage of the mutated components of the neural network matrices.
# must have a value between 0.0 and 1.0
mutation_rate=0.5
# the optimizer for the mutation rate between iterations.
# available values: none|sgd
mutation_rate_optimizer=sgd
# the mutation rate decay as momentum.
# if mutation_rate_optimizer is set to 'none', this value will have no effect.
# must have a value between 0.0 and 1.0.
mutation_rate_momentum=0.01