ml.comet.examples.mnist

Class MnistExperimentExample

java.lang.Object

ml.comet.examples.mnist.MnistExperimentExample

public final class MnistExperimentExample
extends Object

https://github.com/eclipse/deeplearning4j-examples/blob/master/dl4j-examples/src/main/java/org/deeplearning4j/examples/quickstart/modeling/feedforward/classification/MNISTSingleLayer.java A Simple Multi Layered Perceptron (MLP) applied to digit classification for the MNIST Dataset (http://yann.lecun.com/exdb/mnist/).

This file builds one input layer and one hidden layer.

The input layer has input dimension of numRows*numColumns where these variables indicate the number of vertical and horizontal pixels in the image. This layer uses a rectified linear unit (relu) activation function. The weights for this layer are initialized by using Xavier initialization (https://prateekvjoshi.com/2016/03/29/understanding-xavier-initialization-in-deep-neural-networks/) to avoid having a steep learning curve. This layer will have 1000 output signals to the hidden layer.

The hidden layer has input dimensions of 1000. These are fed from the input layer. The weights for this layer is also initialized using Xavier initialization. The activation function for this layer is a softmax, which normalizes all the 10 outputs such that the normalized sums add up to 1. The highest of these normalized values is picked as the predicted class.

Constructor Summary

Constructors

Constructor and Description
MnistExperimentExample()

Method Summary

Modifier and Type	Method and Description
static void	main(String[] args) The experiment entry point.
void	runMnistExperiment(OnlineExperiment experiment) The experiment runner.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

MnistExperimentExample

public MnistExperimentExample()

Method Detail

main

public static void main(String[] args)

The experiment entry point.

You should set three environment variables to run this experiment:

• COMET_API_KEY - the API key to access Comet (MANDATORY)
• COMET_WORKSPACE_NAME - the name of the workspace for your project (OPTIONAL)
• COMET_PROJECT_NAME - the name of the project (OPTIONAL)

Alternatively you can set these values in the resources/application.conf file

Parameters:

args - the command line arguments.

runMnistExperiment

public void runMnistExperiment(OnlineExperiment experiment)
                        throws IOException

The experiment runner.

Parameters:

experiment - the Comet experiment instance.

Throws:

IOException - if any exception raised.