Performance issues with Keras in Regression analysis
I am currently trying to transition my ML code, which we learnt in the first Coursera course as part of deeplearning specialisation, for a fully connected 3 deep neural network. This code was using native Tensorflow and I was using this structure to do a regression analysis.
Since then, I tried transistion this code from native TF to Keras. But I am finding that Keras takes longer and does not converge/reduce loss as quickly as native TF. Does anyone have similar experiences or ideas why this is the case?
Hello @ronyeapen, I recommend you to use Keras API with Tensorflow backend, and I think that it's easier and more practical to implement your code using tf.keras.
Concerning your issue for taking more time of model convergence with Keras, you can accelerate your training process by changing the run-time into GPU, I'm giving you my suggestion based on my experience!