customdl 1.0.14

Custom Deep Learning

Custom Deep Learning

• Create a customized Feedforward Neural Network by changing the number of layers, activation functions, loss function and optimizer.
• Refer to the documentation of any class/method by using help(class/method) Eg: help(FNN), help(FNN.compile)
• For intuitive explanations of the underlying theory refer:
  • Feedforward Neural Network
  • Optimizers

Installation

$ [sudo] pip3 install customdl

Development Installation

$ git clone https://github.com/Taarak9/Custom-DL.git

Usage

>>> from customdl import FNN

Creating a Feedforward Neural Network

# number of input nodes
n_inputs = 27
loss_fn = "ce"
nn = FNN(n_inputs, loss_fn)

# Add a layer with 9 nodes and activation function ReLU
nn.add_layer(9, "relu")
# Add a layer with 3 nodes and activation function sigmoid
nn.add_layer(3, "sigmoid")

# Note the last layer you added will be the output layer of the NN
# Compile the nn
nn.compile(training_data, test_data)

• Available options:
  • Weight initialization: Random, Xavier, He
  • Activation functions: Identity, Sigmoid, Softmax, Tanh, ReLU
  • Loss functions: MSE, Cross Entropy
  • Learning mode: online, mini-batch, batch
  • Optimizers: GD, Momentum based GD, Nesterov accerelated GD
• Refer to the Handwritten digit recognizer built using this package.

To-do list

• Use validation data for hyperparameter tuning
• Plots for monitoring loss and accuracy over epochs
• Regularization techniques: L1, L2, dropout
• Add optimizers: Adam, RMSProp
• RBF NN