Michael Shoemate

Docs Section: How to choose budget parameters

1

What is a reasonable privacy utilization? How do you interpret the risk to individuals for any given budget parameter in any given privacy definition? How to distribute budget parameters across...

Investigate when Python constructors can do dtype coercion

2

Now that dtypes are typically filled in via partial chaining, we may be able to relax the checks on certain int vs float arguments. Integers can be coerced to floats,...

standardize notation for neighboring data sets

2

Closes #1312

There should be a way to add user-defined constructors to the constructors dict in `context.py`. One way could be via the following functions in the extrinsics module (that are not...

prove integer Laplace mechanisms

2

Closes #1313

prove float Laplace mechanisms

2

Closes #1314

adjust behavior of linear discrete laplace sampler for constant time

2

Closes #1324

I think the best place to raise the warning is in `dp.loss_of`, and add another argument to hide warnings (`warn=False`)? Threshold based on rule of thumb: * epsilon no greater...

Context API: edge-case where chaining fails

2

``` Traceback (most recent call last): File "/Users/michael/openDP/openDP/.binder/out/2024/demo-drifting-spaces.py", line 24, in subcontext.query().laplace() File "/Users/michael/openDP/openDP/python/src/opendp/context.py", line 422, in make new_chain = self._chain >> new_chain ~~~~~~~~~~~~^^~~~~~~~~~~ TypeError: unsupported operand type(s) for >>:...

rename `output_measure` to `privacy_measure`

6

There's never a distinction between input vs output measure, so "output" is unnecessary. But there is a distinction between a measure (general mathematics) and a privacy measure (DP), and we...