Hi, now I'm facing new problem at the construction of neural network in tensotflow.
I would like to implement regression/classification problem at a same time in one neural network, which has two outputs, one for the classification [discrete variable as 0 or 1] and the other for the regression [continuous positive variable like time or length]. This kind of network shares any weights and biases for the several outputs.
In the deepnet library, I can construct this kind of network by jointly locating multiple layers (sigmoid and ReLU) for the output like multi-modal network, however, does anyone know how to do it with Tensorflow?
Standard OLS regression is equivalent to minimizing an L2 loss function of a neural network with 1 layer that has the identity as its activation function. Standard Logistic regression is equivalent to minimizing an L2 loss function of a nueral network with 1 layer that has a sigmoid as its activation function.
Refer this article
https://www.datacamp.com/community/tutorials/keras-r-deep-learning
Thank you for paying attention to my question.
> Chris
Thank you for answering. Of course, I know how to construct standard linear or logistic regression on tensorflow and its way of optimization. However, my interest is the construction of DEEP neural network with MULTIPLE OUTPUTS, that is why I called it "multivariate" or perhaps "multimodal".
> Rajani
Thank you for answering. I know the way to create basic neural network models such as Multi-Layer Perceptron to implement regression/classification respectively.
So, what I don't get is the code expression to implement deep learning with several outputs in tensorflow. Thank you.
Thanks to you all, I finally made it to implement multivariate regression on neural network with tensorflow.
[codes on pic 1]
However, the optimization to minimize this summarized overall MSE goes wrong when the scale (variance) of Y1 and Y2 greatly differs. For example, in the result below, Y2 MSE began to increase while Overall MSE and Y1 MSE continuously decrease.
[results on pic 2]
Any good ways to optimize this?
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis