tf.debugging.check_numerics  |  TensorFlow v2.16.1 (original) (raw)

Checks a tensor for NaN and Inf values.

View aliases

Compat aliases for migration

SeeMigration guide for more details.

tf.compat.v1.check_numerics

tf.debugging.check_numerics(
    tensor: Annotated[Any, TV_CheckNumerics_T], message: str, name=None
) -> Annotated[Any, TV_CheckNumerics_T]

Used in the notebooks

When run, reports an InvalidArgument error if tensor has any values that are not a number (NaN) or infinity (Inf). Otherwise, returns the input tensor.

Example usage:

a = tf.Variable(1.0)
tf.debugging.check_numerics(a, message='')

b = tf.Variable(np.nan)
try:
  tf.debugging.check_numerics(b, message='Checking b')
except Exception as e:
  assert "Checking b : Tensor had NaN values" in e.message

c = tf.Variable(np.inf)
try:
  tf.debugging.check_numerics(c, message='Checking c')
except Exception as e:
  assert "Checking c : Tensor had Inf values" in e.message