Summary: 32 instances, 32 unique

Text	Count
# TODO: Test whether adding min/max capping to dykstra projection would	1
# TODO: support multi dim inputs.	1
# TODO: Add final projection for unimodality constraints.	1
# TODO: add support for tf.map_fn and inputs of shape (B, ?, input_dim)	1
# TODO: optimize for case where all dims are monotonic and we won't	1
# TODO: investigate if there is a way to avoid sorting twice.	1
# TODO: investigate whether eps should be bigger.	1
# TODO: It is likely that this algorithm will work for all trapezoid	1
# TODO: Rename and update usage.	1
# TODO: unstacking entire set of weights for the purpuse of projection	1
# TODO: add support for serialization and object scoping or annoations.	1
# TODO: Add support for calibrators with units > 1.	1
# TODO: run benchmark and figure out whether it make sense to merge	1
# TODO: Determine partial ordering of features by dominance and	1
# TODO: can we remove this now that we always project at every step?	1
# TODO: figure out whether it should be used.	1
# TODO: come up with a better solution than separately applying	1
# TODO: support multi dim output.	1
# TODO: in the case of only one monotonic dimension, we only have to	1
# TODO: handle cyclic PWL layers.	1
# TODO: actually assert them.	1
# TODO: If warmstarting, look for the previous ensemble file.	1
# TODO: add option for different pre-aggregation model (linear/ensemble)	1
# TODO: add linear layer regularizers.	1
# TODO: add examples in docs.	1
# TODO: add support for KFL in RTL Layer	1
# TODO: make _REPEATED_PAIR_DISCOUNT_IN_CRYSTALS_SCORE config param	1
# TODO: implement variation/variance regularizer.	1
# TODO: add support for different lattice_sizes for each input	1
# TODO: Add built-in regularizers like laplacian, hessian, etc.	1
# TODO: update library not to explicitly check if None so we can return	1
# TODO: approach used to implement regluarizers is likely to be more	1