Add a pretty formatter for rotation angles so we can have nicer musical score diagrams

[tanujkhattar]

When plotting musical score diagrams for QvrZPow with a symbolic rotation angle sympy.Symbol('Y'), I want the output to have nicely formatted rotation angles instead of very small floating point values. For example:

When running the code

qvr_zpow = QvrZPow(Register('x', QFxp(12, 6, False)), gamma = sympy.Symbol('Y'))
show_bloq(qvr_zpow, 'dtype')
show_bloq(qvr_zpow.decompose_bloq(), 'musical_score')

I want the output to be like

instead of (current main)

or what you get if you update the pretty name of ZPowGate bloq to return f'Z^{self.exponent}'

To generate the original diagram, I implemented a very hacky format function locally as follows:

def format(x: SymbolicFloat) -> str:
    if isinstance(x, float) or x.is_constant():
        if x.is_integer:
            return f'{int(x)}'
        for i in range(1, 20):
            if x * (2**i) == 1:
                return f'/2^{i}'
        return f'{x:.2g}'
    if isinstance(x, sympy.Symbol):
        return str(x)
    if isinstance(x, sympy.Expr):
        return '*'.join(format(x) for x in x.args[::-1])
    return str(x)

The goal of this issue would be to add a similar production quality format function that can be used to get nice string representations of exponents represented as a SymbolicFloat

cc #791

[fdmalone]

if you update the pretty name of ZPowGate bloq to return f'Z^{self.exponent}'

What if you return f'$Z^{self.exponent}$'

[fdmalone]

Oh the issue is fractional formatting not that the label isn't rendered as latex

[mpharrigan]

What does your ideal solution look like here? Would it be mostly solved by #791? Shouldn't sympy be responsible for keeping the expression in fractional form? Remember that there is a big difference between sympify('x^(1/2)') and 'x^(1/2.)'

[jbolns]

Got to this via unitaryHack.

I think I'm getting somewhere, but need to put some additional work formatting the last bit. It's a catch-and-release operation using a combination of RegEx and Sympy. Since there is a timestamp associated with the unitaryHack and I am new to the codebase, this seemed the wise/cautious approach.

Could I ask about any significant changes to the relevant parts of the codebase since the issue was opened?

I had to hack the qualtran/bloqs/rotations/quantum_variable_rotation.py::build_composite_bloq (offset was wrecking chaos) a tiny bit to get the code below working... Although, to be fair, I'm also a little scared to ask as I might have missed something obvious due to time constraints.

qvr_zpow = QvrZPow(Register('x', QFxp(12, 6, False)), gamma = sympy.Symbol('Y'))
show_bloq(qvr_zpow, 'dtype')
show_bloq(qvr_zpow.decompose_bloq(), 'musical_score')

[jbolns]

Yup, looking pretty.

Do you mind massively if I submit the code for review, somehow?

[mpharrigan]

PRs welcome!

[jbolns]

Sweet!

Done ^

On EU time zone, so it's lights off for today. Happy to adapt upon feedback.

[jbolns]

I pushed a second commit to PR.

I sat down to code it again with the benefit of lessons learned (it often helps me produce more refined code). So, not a refactoring, but it still follows a similar rationale: split, evaluate, combine.

Apologies for making you read two files. I don't typically participate in hackathons because I like to re-write things over until I'm happy. I just couldn't resist this time.

Of course, happy for feedback.