-
Notifications
You must be signed in to change notification settings - Fork 0
Home
ANN4j is a java package that provides Object oriented Neural Networks for making Explainable Networks. Object Oriented Network structure is helpful for observing each and every element the model. This package is developed for XAI research and development.
READ COMPLETE DOCUMENTATION HERE
ANN4j is a java package that provides object oriented functionality to neural networks. It implements multilayer perceptrons in java by using Objects instead of matrix multiplications. Every neuron is treated as a seperate object. While this kind of implementation is highly inefficiant when compared to matrix multiplications, this implementation will help research in the fields of Explainable AI. Explainable AI aims at making the model interpretable. By pausing and observing the neural net at different stages, researchers can study neural networks more efficiantly. Indivisual observable interfaces are more easy to observe then matrices. Operations which are difficult to perform on matrices can be performed more easily using this technique.
- Observable implementation for Artificial Neural Networks (ANN)
- XAI method for relevance propagation
- Stochastic/batch gradient descent
- No hardcoded implementations lets researchers change the parameters as they want.
- Plug and play mnist type data. Other Data files can be handeled via extension
The package can be imported after download.
import ann4j.*;
None. This package is made using 100% Pure Java. The java package requires java 5.0+. No other requirements are required. Recommended to use the latest version of Java. Java download link
Exaustive information can be found on our wiki
![image](https://user-images.githubusercontent.com/83284294/208302425-951e57e9-d389-42e1-81e2-aa54148cc21e.png)
![image](https://user-images.githubusercontent.com/83284294/208826148-f679c611-dac5-4f87-997a-cd024f0376b5.png)
- Strategy Pattern - NeuronBehaviour
- Observer pattern - NeuronObserver
- Template pattern - Trainer
- Singleton Pattern - NeuronBehaviour concrete classes
-
Setting the output file to be output.txt and enabling command line logging.
parameter.setOutputFile("output.txt", true);
-
Setting the number of neurons in each layer.
parameter.setLayerArray(784, 32, 16, 16, 26);
-
Setting the training file to be emnist-letters-train.csv and the file type
parameter.setTrainingFileReader("emnist-letters-train.csv", "mnist");
-
Setting the testing file
parameter.setTestingFileReader("emnist-letters-test.csv", "mnist");
-
Setting the learning rate for weights
parameter.setLearningRate(1);
-
Setting the learning rate for the bias to 1.
parameter.setBiasLearningRate(1);
-
Setting the epsillion value for the relevance propagation algorithm.
parameter.setEpsillion(0);
-
Setting the batch size
parameter.setBatchsize(10);
-
Setting the rectification function.
parameter.setRectificationFunction("sigmoid");
-
Creating a new instance of the Trainer class.
Trainer myTrainer = new Trainer();
-
Training the network with 88800 samples for n epochs
myTrainer.train(m, n);
-
Creating a new instance of the NeuronObserver class this class will observe the neurons and respond when every parameter is changed.
NeuronObserver myNeuronObserver = new NeuronObserver();
-
Testing the network with 9990 samples.
myTrainer.test(9990);
-
Adding the neuron at layer 1 and index 31 to be observed.
myNeuronObserver.addNeuronToBeObserved(1, 31);
-
Training accuracy
myTrainer.getModelEvaluator().getTrainingAccuracy();
-
Testing accuracy
myTrainer.getModelEvaluator().getTrainingAccuracy();
-
Confusion Matrix
myTrainer.getModelEvaluator().printConfusionMatrix();
-
xai algorithm for relevance propagation.
myTrainer.relevancePropagate(2, 3);
-
xai algotithm for most significant input neurons
myTrainer.forwardPropagatewithExclusionInputLayerOnKSamples(2);
In ANN4j, every neuron is an object of its own. Every Neuron can be observed by the NeuronObserver class when the values are updated. NeuronObserver class can be extended as per the requirement of the parameters to be observed. Neurons objects can also be obtined and observed independantly.
-
Get a neuron object from a layer.
myTrainer.getLayerManager().getLayer(layerNum).getNeuron(neuronNum));
-
Get activation of a neuron
neuron.getActivation();
-
Get bias of the neuron
neuron.getBias();
-
Get arraylist of the left or right connections of the neuron
neuron.leftConnections; neuron.rightConnections;
-
Get weight of a connection
connection.getWeight();
Output can be seen in the file specified by the parameter class. The write class can be used to write any user defined Strings to the file.
Writer.write();
This is an example of the output generated by training and observing a neuron.
Training accuracy in epoch 0 is 10.66891891891892
Testing accuracy 28.92892892892893
The neuron 31 in layer 1 has been updated by forward propagation
Neuron #31 has activation 6.567825572210979E-4
The neuron 31 in layer 1 has been updated by forward propagation
Neuron #31 has activation 0.003181628304117291
Testing accuracy 28.93314651721377
The neuron 0 in layer 2 has been updated by forward propagation
Neuron #0 has activation 0.25373727956231534
The neuron 0 in layer 2 has been updated by forward propagation
Neuron #0 has activation 0.7220061457959416
Testing accuracy 28.94736842105263
The default format for the package is MNIST format.
File type CSV consisting of the following
- 1 Label (Expected number)
- n pixel weights n must match number of input neurons.
Example
2,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,116,125,171,255,255,150,93,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,169,253,253,253,253,253,253,218,30,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,169,253,253,253,213,142,176,253,253,122,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,52,250,253,210,32,12,0,6,206,253,140,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,77,251,210,25,0,0,0,122,248,253,65,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,31,18,0,0,0,0,209,253,253,65,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,117,247,253,198,10,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,76,247,253,231,63,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,128,253,253,144,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,176,246,253,159,12,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,25,234,253,233,35,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,198,253,253,141,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,78,248,253,189,12,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,19,200,253,253,141,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,134,253,253,173,12,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,248,253,253,25,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,248,253,253,43,20,20,20,20,5,0,5,20,20,37,150,150,150,147,10,0,0,0,0,0,0,0,0,0,248,253,253,253,253,253,253,253,168,143,166,253,253,253,253,253,253,253,123,0,0,0,0,0,0,0,0,0,174,253,253,253,253,253,253,253,253,253,253,253,249,247,247,169,117,117,57,0,0,0,0,0,0,0,0,0,0,118,123,123,123,166,253,253,253,155,123,123,41,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
![image](https://user-images.githubusercontent.com/83284294/208296890-8655f164-ffe5-41b3-865d-c28c909ded2c.png)
Here the image 2 is represented as an array of 28*28 pixels each value represents pixel activation.
Some datasets to test the package on (without extending mnist file reader)
- MNIST Handwritten database
- MNIST extended chracter database and Kaggle link
- MNIST fashion data set
- Kannada MNIST
References
ANN4j provides functionality to extend the InputFileReader to add file handling for various types of datasets apart from mnist type files. InputFileReader or MNISTFileReader can be extended by making relevant changes in file reading functions.
The new file reader class must pass the filename to the super constructor.
super(filename);
The next()
method is responsible for reading new line input from the dataset. It must also act as a super setter method. It must set all values like label
, expectedOutputArray
and inputArray
public void next()
This method must return the label (expected value of prediction).
public double getLabel()
This method must return the values of neurons (expected value of prediction). Example for digit recognition fo digit two, the arraylist must contain 784 elements of the pixel values.
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,116,125,171,255,255,150,93,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,169,253,253,253,253,253,253,218,30,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,169,253,253,253,213,142,176,253,253,122,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,52,250,253,210,32,12,0,6,206,253,140,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,77,251,210,25,0,0,0,122,248,253,65,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,31,18,0,0,0,0,209,253,253,65,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,117,247,253,198,10,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,76,247,253,231,63,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,128,253,253,144,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,176,246,253,159,12,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,25,234,253,233,35,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,198,253,253,141,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,78,248,253,189,12,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,19,200,253,253,141,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,134,253,253,173,12,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,248,253,253,25,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,248,253,253,43,20,20,20,20,5,0,5,20,20,37,150,150,150,147,10,0,0,0,0,0,0,0,0,0,248,253,253,253,253,253,253,253,168,143,166,253,253,253,253,253,253,253,123,0,0,0,0,0,0,0,0,0,174,253,253,253,253,253,253,253,253,253,253,253,249,247,247,169,117,117,57,0,0,0,0,0,0,0,0,0,0,118,123,123,123,166,253,253,253,155,123,123,41,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
public ArrayList<Double> getInputArray()
This method must return the expected values of output neurons (expected value of prediction). Example for digit recognition fo digit two, the arraylist can be
0,0,1,0,0,0,0,0,0,0
This is dependant on the model and is a design decision.
public ArrayList<Double> getExpectedOutputArray()
This method is used for obtaining the prediction value from the activations in the output neurons. In the digit recognitoin case as every input is mapped with same neuron it is the same. For example if neuron number 3 fires the highest, the model has predicted 3. But this needs to be overridden for different model configurations.
public double getPredictionFromNeuronNum(int mostSignificantNeuronNumAsPrediction)
Note-
Predicted neuron and prediction are different.
Predicted neuron is the neuron which is most significant in firing. The prediction is the value corresponding to that neuron.
Example Consider case of handwritten letters database. If the neuron 4 is most significant (glows brightest) and it corresponds to label D then the predicted neuron is 4 and prediction is D.
getMostSignificantNeuronNumAsPrediction()
is a method in LayerManager class which helps to get the value of the neuron which fires the most.
Creates a new instance of the file reader and starts all over again.
public void restart()
After the file reader custom class had been made, it can be passed to the parameter class using the methods
public static void setTrainingFileReader(InputFileReader inputFileReader);
public static void setTestingFileReader(InputFileReader inputFileReader){
Please visit the documentation wiki page https://aatmaj-zephyr.github.io/ANN4jwiki/
-
Example code https://github.com/Aatmaj-Zephyr/ANN4j/blob/main/Main.java
-
Sample output https://github.com/Aatmaj-Zephyr/ANN4j/blob/3721148ec24371bf095e1394fe39fc471f391466/output.txt
-
Sample output https://github.com/Aatmaj-Zephyr/ANN4j/blob/ef0f34b505e6e6316f94b5a660b9ef651582667d/output.txt
- More about Artificial Neural Networks https://www.3blue1brown.com/topics/neural-networks
- Relevance propagation example https://towardsdatascience.com/indepth-layer-wise-relevance-propagation-340f95deb1ea
- Rectification functions https://www.quora.com/What-is-the-purpose-of-rectifier-functions-in-neural-networks
Please feel free to suggest any changes or point out any errors by raising an issue here
For asking for clarification on any topic, raise an question issue here
Please read the contributing guidelines here. Everyone is free to contribute to this project.
Are you using ANN4j in your research or project? If so, please let me know and I may add a link to your project or application and your logo to this repository. Also please consider starring this repository and following me.
You can cite this repository using the following bibtex entry. Please update the date.
@misc{AatmajZephyr21:online,
author = {Aatmaj Mhatre},
title = {Aatmaj-Zephyr/ANN4j: Artificial Neural Networks for Java This package provides Object oriented Neural Networks for making Explainable Networks. Object Oriented Network structure is helpful for observing each and every element the model. This package is developed for XAI research and development.},
howpublished = {\url{https://github.com/Aatmaj-Zephyr/ANN4j}},
month = {},
year = {},
note = {(Accessed on <date>)}
}
MIT License
Copyright (c) 2022 Aatmaj
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
Thank you For visiting. Please share your love by starring this repository and following me!