Summary: 60 instances, 57 unique

Text	Count
AttributeValue = Any  # TODO Union[Sequence[float], Sequence[int], Sequence[Text], Sequence[TensorProto], Sequence[GraphProto]]	1
// TODO: check that the index is valid given the length of the input array.	1
# TODO: There's no MIL op to extract a value from a symbolic tensor,	1
# TODO: remove InternalVar check	1
// TODO -- This seems like a giant hack. This is leaking abstractions.	1
// TODO: this should do something.	1
# TODO: If input is not rank 3 or 4, then accordingly handle	1
# TODO: The calculations to convert adaptive_pool to standard pool,	2
# TODO: Skip or CopyProp + DeadCode elimination	1
# TODO: Add one expand instead of adding one after another for input, h and c	1
# TODO: instead of a hard-coded set, use op-traits	1
TODO (rdar://59165842): Use expand_dims, squeeze etc to use 0 instead	1
# TODO: Add one expand instead of adding one after another for input, h	1
// TODO @znation: commenting out the below, because it breaks pipeline validator.	1
// TODO: Compare sizes	1
# TODO: All outputs that depend on only invariants are invariant. We	1
# TODO 3D convolution doesn't support dynamic weights:	1
// TODO: make this check all of the blob names, and make it compare to the input and output shapes	1
## TODO: Add un-supported ops	1
# TODO: if it's an embedding, this should be integer	1
# TODO: Dynamic rank: Use GetShape and select indices dynamically	1
// TODO -- is there a better way to do this than switch/case?	1
# TODO: Right now, "const elimination" pass CANNOT be done after the "homogenize_input_dtypes" pass.	1
# TODO: gather doesn't work when the shape is known size.	1
// TODO @znation: validate array length below	1
// TODO: Check parameters match	1
// TODO: there is some cross shape constraint here -- all 3 or 4 input ranges might be unconstrained, but now they are constrained to a combination whose product is	1
# TODO Should we support this case?	1
## TODO: Test coverage when B and C are non-constant	1
// TODO -- validate classifier interface	1
// TODO: add check that this size is possible from the input size	1
# TODO: ONNX should be able to infer the shape	1
// TODO put try/catch around this?	1
// TODO: Check to make sure the particular option is compatible with a classifier	1
# TODO: Replace If -> ElIf with more general transformation block	1
// TODO: add a catch with an error message that mentions the name	1
## TODO: rdar://73851694 (Update einsum op translation to support generic cases)	1
* TODO: rdar://76017556	1
WRAP_SIMPLE_TYPE(Image, MLFeatureTypeType_imageType) /* TODO image is not simple type */	1
// TODO: need to forward the other constraints -- sequence and batch (although those can't exist for this layer)	1
# TODO : If tf.split output is returned, there's no	1
// TODO Probability outputs are always dictionaries!	1
# TODO: extract nodes that depends on the "const part" of placeholders.	1
// TODO -- validate regressor interface	1
# TODO: Adding Linear layer will change to	1
// TODO: make enumgen generate this when compiling .proto files	1
// TODO: Error checking routines -- make sure this node hasn't been requested before.	3
# TODO: rdar://65575826 (PyTorch converter: output_padding mapping to slice	1
# TODO: To be removed once, auto-downgrading of spec version is enabled	1
## TODO: Add support for conversion from STRING TO FLOAT	1
# TODO: Skip or CopyProp + DeadCode elimination	1
# TODO (sberardi): 3rd param to aten::sub is a scale factor, need to handle that.	1
TODO:	1
{Specification::FeatureType::TypeCase::kMultiArrayType, // TODO ARRAY TYPE IS INVALID, REMOVE	1
// TODO -- not implemented!	1
# TODO (sberardi): 3rd param to aten::add is a scale factor, need to handle that.	1
# TODO: Move error check under VERBOSE / DEBUG Mode	1