Add ROIFilter #173
Add ROIFilter #173
Conversation
| counts = data.bins.size().to(dtype='int32', copy=False) | ||
| counts.unit = 'counts' | ||
| self._add_counts(counts) | ||
|
|
||
| def add_counts(self, data: Sequence[int]) -> None: |
There was a problem hiding this comment.
This will not be used by Beamlime anymore, keeping it for backwards compatibility.
| size = _noise_size | ||
| position = sc.empty(sizes={'position': size}, unit='m', dtype=sc.DType.vector3) | ||
| position.values = np.random.default_rng().normal(0, sigma.value, size=(size, 3)) | ||
| position.values = np.random.default_rng(seed=1234).normal( |
There was a problem hiding this comment.
Is this to turn the centers of the pixel positions into small gaussian clouds of points?
I was thinking maybe it should be random.uniform instead to avoid spill-over from one pixel over to the next?
But maybe using a normal distribution is better as it makes the clouds spherical, and they then look the same from any line-of-sight? Using uniform also suggests that the pixels are square in shape, but this is not really true for tubes...
Is the sigma approximately equal to half the width of the pixel?
There was a problem hiding this comment.
The gaussian option is just a simple alternative to something that uses the pixel shape, see the cylinder case (used for Loki tubes) which is uniform within the pixel. We may eventually have something similar for other shapes, I think.
Sigma is a parameter than can be set in a config file in Beamlime.
src/ess/reduce/live/roi.py
Outdated
| indices will be returned concatenated into a dense array. | ||
| """ | ||
| out_dim = 'index' | ||
| for dim, (low, high) in intervals.items(): |
There was a problem hiding this comment.
At some point, we may be hooking this up to a graphical interface where a user will draw a rectangle by hand.
In some plotting libraries (e.g. MPL), when you request the bounds of a rectangle, the order of the bounds depends on how the user drew the rectangle.
If they started from the left and dragged towards the right, you get [left, right] as the bounds.
They they started from the right, you get [right, left] instead.
The same goes fro top/bottom.
We could end up in a situation where some of the tuples contain 2 values but the second one is lower than the first. The slicing below would then select nothing.
I recommend we add something like low, high = sorted(low, high) to make sure that does not happen further down the line.
There was a problem hiding this comment.
Good point, done, please have another look.
This will be used by Beamlime to make time-of-arrival histograms of counts within an ROI. Currently this only supports axis-aligned rectangle selections.