DEPTH

A DEep learning PyTHon library written using numpy only.

Usage Instruction

• Clone the git repo using

  git clone https://github.com/diwasblack/depth.git
  

• Activate the virtual environment if needed
• Install library from the root folder in editable mode using

  pip install -e .
  

Features

• Dense layer with activations:
  • Sigmoid
  • Tanh
  • ReLu
  • LeakyRelu
  • Softmax
  • Linear
• Convolution layer with activations:
  • ReLu
• Pooling layer
  • Max pooling
• Optimizers
  • Stochastic Gradient Descent
  • ADAM
• Loss functions
  • Mean Squared Error
  • Cross entropy loss
• Regularizer
  • L2 Regularizer

Examples

A sequential network with 2 hidden layer and a softmax layers as final layer with cross entropy loss.

nn_object = Sequential()
nn_object.add_layer(DenseLayer(
    units=32, activation="tanh", input_dimension=10))
nn_object.add_layer(DenseLayer(
    units=64, activation="tanh"))
nn_object.add_layer(DenseLayer(units=10, activation="softmax"))
nn_object.compile(loss="cross_entropy", error_threshold=0.001)

See the examples folder for more.