How do you debug and construct the model by applying the tensorflow?

[mtli77]

Hi @Yang7879
Thank you very much for sharing the great project!
I am new to tensorflow. However, I found it quite difficult to debug on tensorflow, which means that with the sess.run(), I can not see the real training data transform in the model.
Could you please share me some experience on how to build the model correctly, and how to debug with this static graph framework effectively?
Waiting for your reply~~~

[Bo Yang]

Hi @Violetit , I understand that TensorFlow is a bit painful for beginners, but after one or two months, you will feel it's pretty easy and flexible to use. Here are two important tips:

1. Define the input and output variables at the beginning, and treat them as a batch of data points.
2. Once you design the neural layers, you need to always track and check the shape of all tensors, confirming the shape is correct before you design another layer or operation. I recommend using the debug button of Pycharm.

Actually there's no need to feed data to debug, it's meaningless for you to check the numbers because you don't understand them anyway.