Deep learning Java library

As we’ve heard probably there is a ML toolkit based on Java, - DL4J from Eclipse. But now we have another one library for ML on Java from AWS team. And it’s good.

I can watch endlessly dancing graphs. But let’s see, what we have.

Amazon released Deep Java Library, an open-source library for training, testing, deploying, and making predictions with deep-learning models.

There are jupyter’s notebooks here. Maven:

<!-- https://mvnrepository.com/artifact/ai.djl.mxnet/mxnet-engine -->
<dependency>
    <groupId>ai.djl.mxnet</groupId>
    <artifactId>mxnet-engine</artifactId>
    <version>0.2.0</version>
</dependency>

Deep Java Library supports training on multiple GPUs.

TrainingConfig config =  new DefaultTrainingConfig(initializer, loss)
    .setOptimizer(optimizer)
    .setBatchSize(batchSize)
    .setDevices(Device.getDevices(1))
    .addEvaluator(accuracy);

We can use setDevices and pass an array of devices we want the model to be trained on. For example, new Device[]{Device.gpu(0), Device.gpu(1)} for training on GPU0 and GPU1.