Summary: 34 instances, 32 unique

Text	Count
/* TODO - skip this logic for now. It requires walking back the inputs which is not yet possible	1
// TODO output = backtrackIdentity(output); doesn't seem to work with Java version.	2
/* TODO need ability to walk back inputs	1
// TODO output = output.op.inputs[0]	1
// TODO output = backtrackIdentity(output);	1
// FIXME need to manage input/output operand lifetimes	1
// TODO  if( output.op().numOutputs() != 1)	1
// TODO use in generator?	1
// TODO   if (output.op().numInputess() != 1)	1
// TODO   output = output.op().inout(0);	1
// TODO     throw new IllegalArgumentException("output can only have 1 output");	1
// TODO make this be used.	1
// TODO handle Ragged Tensors????	1
// TODO: hold for lazy select feature,	2
// TODO - this does not work	1
// TODO should this be moved to FramnworkOps under math.	1
// TODO Python Keras returns a "grad", which is an optimization not implemented in Java.	1
// TODO add attribute type once supported	1
// TODO should pass the scope instead	1
// TODO get attribute names.  Needs TF 2.7	1
// TODO if (output.op().type().equals(Sigmoid.OP_NAME)) {	1
* FIXME: This causes native errors when I use it (Linux GPU, 6.1 CC), but I'm leaving it because	1
// TODO add Func once supported	1
// TODO handle resource variables, too	1
// TODO optional function attrs (there currently aren't any)	1
// TODO   return tf.nn.sigmoidCrossEntropyWithLogits(target, output);	1
// TODO (karllessard) scan and registered automatically all annotated tensors types	1
// TODO (karllessard) could be supported by adding this attribute type in the eager C API	1
// TODO this was tf.math.l2_normalize in TF Python	1
// TODO better way of doing this, only count as ran assignments to the restored variables.	1
// TODO handle ragged Tensors	1
.startsWith("_"); // TODO do I want this?  Some interesting ops like _XlaCompile	1