Printing tensor values with tf.Print()

Another option to print values for debugging purposes is to use tf.Print(). You can wrap a tensor in tf.Print() to print its values in the standard error console when the path containing the tf.Print() node is executed. The tf.Print() function has the following signature:

tf.Print(
     input_,
     data,
     message=None,
     first_n=None,
     summarize=None,
     name=None
    )

The arguments to this function are as follows:

• input_ is a tensor that gets returned from the function without anything being done to it
• data is the list of tensors that get printed
• message is a string that gets printed as a prefix to the printed output
• first_n represents the number of steps to print the output; if this value is negative then the value is always printed whenever the path is executed
• summarize represents the number of elements to print from the tensor; by default, only three elements are printed

Note

You can follow along with the code in the Jupyter notebook ch-18_TensorFlow_Debugging...