/deeplearning_scala

scala实现的深度学习deeplearning,包括da,rbm,sda,dbn,cnn,rnn

Primary LanguageScala

deeplearning_scala

deeplearning with scala(2.10.6) java-SE 1.8

  • deeplearning with scala and spark
  • Copyright liming(oldlee11)
  • Email: oldlee11@163.com
  • qq:568677413

<一> src/dp_process/* include :(串行基本方式实现)

LogisticRegression(逻辑回归层softmax输出)

HiddenLayer(常规神经网络的隐藏层)

Dropout(多层感知器MLP=多个HiddenLayer+LogisticRegression)

dA(对Hiddenlayer的单层预训练dA方式)

sdA(对MLP的多层预训练dA方式)

RBM(对Hiddenlayer的单层预训练RBM方式)

DBN(对MLP的多层预训练RBM方式)

ConvPoolLayer(卷积层+max池化层)

CNN(卷积神经网络=多个ConvPoolLayer+Dropout)

RNN(多层卷积升级网络=多层RNN_HiddenLayer+1层RNN_LogisticRegression,没有实现lstm,支持多层的mlp的递归)

CNN+dA(卷积神经网络+dA方式预训练[不确定是否完全正确])

使用minst数据集做实验

sdA:88.17%

DBN:88.7%

CNN(lenet5):90.3%

<二> src/dp_process_parallel/* include :(并行方式实现)

Dropout_parallel=多个HiddenLayer_parallel+LogisticRegression_parallel

CNN_parallel=多个ConvPoolLayer_parallel+Dropout_parallel

使用minst数据集做实验

Dropout(mlp):86.22%

CNN(lenet5):96.08%

<三> src/dp_process_breeze/* include :(breeze+并行方式实现)

   ..使用breeze blas jblas包