Summary: 31 instances, 30 unique

Text	Count
-- TODO: Look into hs-boot refactoring to allow mutually recursive imports.	1
-- TODO: Support heterogeneous list of tensors.	1
-- TODO: better name.	1
encodeTensorData = error "TODO (Complex Double)"	1
-- TODO: strictness (e.g., foldlM')	1
-- TODO: given that distinction, figure out a good story around	1
-- TODO: Address 1D convolution.	1
-- TODO: unify with BuildResult (Tensor v)	1
# TODO: remove this once we stop depending on `snappy`.	1
tensorVal = error "TODO (Complex Float)"	1
-- TODO: Consider supporting this better in op generation.	1
decodeTensorData = error "TODO (Complex Double)"	1
-- TODO (jcberentsen): Python implementation uses set_shape for	1
-- TODO: be more type safe.	1
-- TODO: Currently, it just makes a copy of the Tensor (and then deletes it),	1
-- TODO: Haskell and tensorflow use different byte sizes for bools, which makes	1
-- TODO: implement support for static shape inference	1
-- TODO: precache the rendering?	1
tensorVal = error "TODO Variant"	1
encodeTensorData = error "TODO (Complex Float)"	1
-- TODO: Generalize for user libraries by replacing getAllOpList with	1
-- TODO: This could be either Int32 or Int64.	2
-- TODO: more generic way	1
-- TODO: Split Data.Digest.CRC32C out of snappy-framing for a	1
-- TODO: Add support for Complex Float/Double.	1
-- TODO: check against existing nodes for conflicts?	1
decodeTensorData = error "TODO (Complex Float)"	1
-- TODO: nicer	1
tensorVal = error "TODO (Complex Double)"	1
-- TODO: Support other convolution parameters such as stride.	1