Bound stream computed field with physical limits

[t-sasatani]

• Sets limits on the computed ADC voltage value.
• The computed voltage is set to zero -1 if it exceeds the limits.
• This doesn't affect the raw metadata value.

---

📚 Documentation preview 📚: https://miniscope-io--82.org.readthedocs.build/en/82/

[coveralls]

Pull Request Test Coverage Report for Build 12058948228

Details

• 6 of 6 (100.0%) changed or added relevant lines in 1 file are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage increased (+0.07%) to 77.286%

---

Totals
Change from base Build 11828727123:	0.07%
Covered Lines:	1065
Relevant Lines:	1378

---

💛 - Coveralls

[sneakers-the-rat]

seems fine to me, i guess my question is when does this occur? seems like if the device is giving us impossible values that's a problem with the device not the i/o software, and setting this to 0 could mask that problem. i trust ya but can i have it explained to me why it's preferable to have 0 in the recorded values rather than the value itself, which can be clamped in analysis?

[sneakers-the-rat]

maybe use -1 to make it unambiguous from the value actually being zero (which should never happen, but thinking about downstream data analysis)

[sneakers-the-rat]

Also I think we are getting to the point where we want to have a separate ScaledValue model, having a bunch of fields named by convention isn't great and not very DRY.

like

T = TypeVar("T", bound = Union[int, float])

class ScaledValue(BaseModel, Generic[T]):
    factor: T = 1
    maximum: Optional[T] = None
    minimum: Optional[T] = None

    def scale(self, value: T) -> T:
        return np.clip(value * factor, self.minimum, self.maximum)

So then for example

class StreamConfig(BaseModel):
    battery_voltage: ScaledValue[float] = ScaledValue(factor = 10, maximum = 20)
    vin_voltage: ScaledValue[float] = ScaledValue(factor=1000, maximum = 20)

where idk we could compute that ScaledValue's factor param from other values like the bit depth if we want to or else make a subclass of ScaledValue.

battery_voltage:
  factor: 100
  maximum: 10
# ...

cfg = StreamConfig()
cfg.battery_voltage.scale(100)

[coveralls]

coverage: 79.162% (+0.05%) from 79.112%
when pulling 42c564f on bound-computed-field
into 56397e9 on main.

[t-sasatani]

[Notes for whenever I can come back] Not done yet. Doing this for every header input feels like a waste and also makes the header-filling process inconsistent. So, it is probably better to define a helper class that is set once with the device config and converts StreamBufferHeader into scaled values.

[t-sasatani]

i guess my question is when does this occur? seems like if the device is giving us impossible values that's a problem with the device not the i/o software, and setting this to 0 could mask that problem.

This mostly happens when the channel or DAQ hardware introduces some bit errors. We're using uint32 to send 8-bit ADC values to simplify the buffer structure, so any flip in the higher bits messes up the metadata plot. I'll change the mask value to -1.