deepdive

deepdive - is a deep learning library implemented in Scala, it uses ND4j for linear algebra.

Layers

• Linear: Linear

• Convolutional: Convolutional

• Embedding: Embedding

• RNN: Vanilla RNN

• LSTM: LSTM

Optimizers

• SGD: the stochastic gradient descent
  • lr: the learning rate, defaults to 0.01
• Momentum:
  • lr: the learning rate, defaults to 0.1
  • momentum: decay parameter, defaults to 0.9

Examples

Fully Connected example

val dataSet = new DataSet(features, labels)

val batchSize = 2
val loss = SoftMaxLoss
val lr = 1e-3

val model = Sequential()
      .add(Linear(3, 3, activation = ReLU))
        .add(Linear(3, 3, activation = ReLU))
          .add(Linear(3, 10))
      .build(loss, Optimizer.SGD(1e-3), batchSize, seed=Some(1))

model.fit(dataSet)

Work in progress!