Controlling the parameters W ,in Chain(Dense(),Dense()) neural network

[yewalenikhil65]

Hello,

I wish to control the behaviour of model while training , based on a constraint on parameters that,

All the parameters (all elements of weight matrix W ) in input layer should stay positive while training the model.

Is this possible currently using Flux ?

[CarloLucibello]

This is similar to #942

I suggest you create a custom dense layer, where you reparametrize the weight as W = exp.(A), where A contains the parameters you actually train

[yewalenikhil65]

Hi,
I was referring to more like following

https://github.com/keras-team/keras/blob/master/keras/constraints.py

which is explained here
https://machinelearningmastery.com/how-to-reduce-overfitting-in-deep-neural-networks-with-weight-constraints-in-keras/

[CarloLucibello]

closing as similar to #1389