Deep Neural Net for Regression
1 view (last 30 days)
I am training a NN with 19 inputs and two outputs. I have 1000 random observations, and I am using the first 800 for training, then 100 additional for validation and the remaining for testing. I am using three hidden layers with 15 nodes each and using tanh for actiation function. I have tried to duplicate all the parameters similar to a Python version. The Python version quickly converges to a MSE of about 0.03. However, the Matlab version is giving a MSE around 10 times the error in Python. I have used every parameter same (min batch size = 32, solver = adam, Learning Rate constant at 0.001 etc. Any help is appreciated. The question is why I am not able to train to the same degress of accuracy. I can provide the scripts and data if any one wants.