tips.md

Tips

Use functions from tf if possible

Math operations on np is not stable for now. Try those alternatives in tf if possible.

For example, use this:

tf.matmul();

...rather than this:

np.matmul();

Use np if saving/loading Tensor

Use the latest TF.NET version

Since TF.NET version correlates that of native TensorFlow, which in turn links to the version of CUDA, it is inconvenient. Though, it is a mission impossible for the small team to maintain all the TF.NET versions. Please use the latest package if possible.