synadart 0.4.1+1 copy "synadart: ^0.4.1+1" to clipboard
synadart: ^0.4.1+1 copied to clipboard

Published 4 years ago • verified publishervxern.dev• Latest: 0.5.0
outdated
→

Metadata

Library for creating neural networks, with a purely Dart implementation.

More...

0.1.0 #

  • Added Multi-layer Perceptron and a basic algorithm for backpropagation

0.1.1 #

  • Added README.md, updated formatting

0.2.0 #

  • Added FF ( feedforward ) and simple Perceptron networks
  • Added LReLU, eLU and tanh activation function
  • Renamed 'sigmoid' to 'logistic' function

0.2.1 #

  • Removed FF ( feedforward ) and simple Perceptron networks in favour of an upcoming simpler implementation of basically the same idea, through just one network model.
  • Added [learningRate] as a parameter, and removed the hard-coded value of 0.2.
  • Organised the files slightly
  • Updated documentation of Neuron

0.2.2 #

  • Updated documentation of Layer and removed a chunk of dead code.

0.2.3 #

  • Updated documentation of Network.
  • Replaced process() in Layer with an output getter, simplifying the implementation of getting each Neuron's output.

0.2.4 #

  • Updated documentation of activation.dart, having added explanations for the different activation functions.

0.2.5 #

  • Renamed Multilayer Perceptron to Deep Feed-forward, which should be a more fitting and future-proof name.

0.3.0 #

  • Updated documentation of Logger, Backpropagation and ValueGenerator.
  • Created examples folder with a recognise_5 example that allows for recognition of the number '5'.

0.3.1 #

  • Added 5 new activation functions: SeLU, Softplus, Softsign, Swish and Gaussian.
  • Renamed the logistic function to sigmoid.
  • Added abs function for obtaining the absolute of a value to mathematical_operations.

0.3.2 #

  • Added simple feed-forward network model.

0.4.0 #

  • Code organisation.
  • Replacing network types such as feed-forward or deep feed-forward with a simple Sequential.
  • Moved focus from Network to Layer, so that different layers can be added to a Network, rather than creating new types of networks, and limiting the user to a preset model.
  • Updated example code and README.md

0.4.1 #

  • Updated documentation

0.4.1+1 #

  • Refactoring code
  • Removing logger.dart in favour of Sprint

← Metadata

15
likes
0
points
24
downloads

Publisher

verified publishervxern.dev

Weekly Downloads

Metadata

Library for creating neural networks, with a purely Dart implementation.

Repository (GitHub)
View/report issues

License

unknown (license)

Dependencies

sprint

More

Packages that depend on synadart

Back