working on general model refactor

[tlambert03]

this PR is not necessarily for merging, but rather for discussing high level model and control flow. I want to step back and reconsider the model carefully. ndv is "doing a thing" somewhat nicely, and it's a good time to step back and clarify exactly what that thing is, with an eye towards building a better foundation to fix some issues (i.e. channel-specific lut persistence) and prepare for some nice features (i.e. chunked loading #22)

cc @alisterburt
@hanjinliu, inasmuch as you're looking at this repo these days, would love your feedback!

Main players here are:

• ArrayDisplayModel: the model representing instructions for visualizing some data source
• LUT: instructions for displaying a specific channel
• ArrayViewer: the view; stubbing out what should be done in response to various events
• DataWrapper: a place to put all of the things we need from the data (we wrap provided data in a DataWrapper based on the type of data).

I'll keep tweaking and updating this flow here, and then actually implement it later

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 74.27%. Comparing base (4e0c5fe) to head (d536b59).

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #33      +/-   ##
==========================================
- Coverage   81.95%   74.27%   -7.69%     
==========================================
  Files          13       13              
  Lines        1236     1236              
==========================================
- Hits         1013      918      -95     
- Misses        223      318      +95     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[hanjinliu]

I think autoscale should be an instance of something like AutoscalePolicy class because There are many possible ways to autoscale an image (maybe even more complicated in the future).

class AutoscalePolicy(ABC):
    @abstractmethod
    def get_limits(self, image: np.ndarray) -> tuple[float | None, float | None]:
        pass

class NoAutoscale(AutoscalePolicy):
    def get_limits(self, image: np.ndarray) -> tuple[float | None, float | None]:
        return None, None

class MinMaxAutoscale(AutoscalePolicy):
    def get_limits(self, image: np.ndarray) -> tuple[float | None, float | None]:
        return image.min(), image.max()

class QuantileAutoscale(AutoscalePolicy):
    lower: float
    upper: float

    def get_limits(self, image: np.ndarray) -> tuple[float | None, float | None]:
        return np.quantile(image, [self.lower, self.upper])

# etc.

[tlambert03]

i like it 👍

[hanjinliu]

Hi @tlambert03, thanks for tagging me. I think it's a nice idea to reconsider the class designs at this point. I'll look into this PR in detail little by little, and leave some comments if I found something might be relevant.

[alisterburt]

I love what I'm seeing at first pass, feels very nice but I need some time to integrate the whole model - need to think a little more on the index/bounds and how this might work/not work with oblique planes... interesting to think about!