Chainer-Python-Logistic回归

Question

I created a simple Logistic Regression model using Python and Chainer but I am not fully satisfied with the end result. Therefore, I would like to get some help. One restriction: I would not like to interchange the implemented functionalities by already existing functionalities. I know that there are loss functions in Chainer which archieve almost the same, but a more complex model I am creating is using a custom loss function. The code is found here:

https://gist.github.com/kmjjacobs/62fc96ece695b47af8d667b060a64559

I would like to keep the model code as clean as possible, but as you can see, the call method is a forward to the loss method and I suspect there is a cleaner way to invoke the loss method in the training loop. I thought that it would be cleaner if the call method outputs the prediction and there is a seperate loss method for computing the loss. What are your thoughts on this?

I am also not sure on the converter function. Is there a better way to achieve the same result?

Do you have any remarks or best practices for writing Chainer code?

Thanks in advance!

Answer 1

At first what is your main question? The best way to define loss function and predict function separately?

I looked your code, I think functionality of init_scope is different between Link and Chain . You cannot use it to register learnable parameter in Chain for this purpose. (Your current usage is for the Link and not for the Chain .)

• init_scope in Link is used to register parameter, https://docs.chainer.org/en/stable/tutorial/function.html#links-that-wrap-functions https://github.com/chainer/chainer/blob/master/chainer/link.py#L197

• init_scope in Chain is used to register other links , https://docs.chainer.org/en/stable/tutorial/basic.html#write-a-model-as-a-chain https://github.com/chainer/chainer/blob/master/chainer/link.py#L675

In your case, I think you can just use chainer.links.Linear to your LogisticRegressionModel , or you can define your own Link class which has the learnable parameter W and use this own link class in your LogisticRegressionModel .