Custom Deep Learning
Project description
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:
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:
- 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
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
customdl-1.0.10.tar.gz
(8.8 kB
view hashes)
Built Distribution
Close
Hashes for customdl-1.0.10-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | f5de725df72122484ed5591447f5f7e8ba2b2ad79e2d666d91e1dd75454fefb6 |
|
MD5 | 0d6d8301260a89cd22ec2b6483eb6913 |
|
BLAKE2b-256 | c114ad02ffb326437d57f6c9806470f17ac69f5010d331bc38382c4d6f4fd550 |