Private Functions in NeuralNetwork

See original GitHub issue

So I have been thinking about https://github.com/BrainJS/brain.js/issues/99 and I think another thing that can get new users bogged down is how many methods and variables are exposed. For example in the REPL I instantiate a simple NeuralNetwork and her are the things I have control over:

How many of these do we actually want the user to be interacting with? If the user trains a network, then randomly calls initialize with some given size their network is dust in the wind.

I think if we take the methods that are meant to only be used internally within the class and declare them in the constructor (or some other scheme for having them not public). I think that might also help those new to the library. I don’t mind doing this If you want me to.

Just check boxes you want Private and I will take care of it.

adjustWeights
calculateDeltasLeakyRelu
calculateDeltasRelu
calculateDeltasSigmoid
calculateDeltasTanh
createTrainStream
formatData
fromJSON
initialize
run
runInputLeakyRelu
runInputRelu
runInputSigmoid
runInputTanh
setActivation
test
toFunction
toJSON
train
trainPattern
activation
biases
binaryThresh
calculateDeltas
changes
deltas
errors
hiddenLayers
hiddenSizes
learningRate
momentum
outputLayer
outputs
runInput
sizes

Issue Analytics

State:
Created 6 years ago
Comments:8 (8 by maintainers)

Top GitHub Comments

3reactions

freddyCcommented, Jan 18, 2018

So I was talking to @robertleeplummerjr about this a bit. The main thing I think would help new users is a more clear interface on which methods are meant for them to interact with. At the same time, I think for those looking through the code it would be more clear to have smaller functions in the neural-network itself, so it is easier to read and understand.

How about adding _ before methods that are meant to be private?

This would let us break out private functions into smaller and easier to read methods, without confusing those starting with the library, since the publicly intended functions are a bit more explicit.

Thoughts?

0reactions

freddyCcommented, Jan 30, 2018

Handled in https://github.com/BrainJS/brain.js/pull/137

Top Results From Across the Web

Private Collaborative Neural Network Learning

To provide a solution to the problem of private collaborative neural network learning, we combine several technologies including MPC, DP, and secret sharing. ......

Private and Reliable Neural Network Inference

Reliable neural networks (NNs) provide important inference-time reliability guarantees such as fairness and robustness. Comple- mentarily, privacy-preserving ...

SecureNN: Efficient and Private Neural Network Training

Neural Networks (NN) provide a powerful method for machine learning training and pre- diction. ... and the output prediction, private from other parties....

Neural Network Tutorial - Ch. 3 Private variables - YouTube

In this chapter we will define all necessary private variables for the network.

Differentially Private Self-normalizing Neural Networks For ...

Differential Privacy involves the addition of random noise to a data sample to produce anonymity.