Add sources and example notebook for 1D CES inference #161

yuema137 · 2024-05-01T22:27:01Z

Note: This PR doesn't affect existing functions in Alea but only adds 1D CES inference.

This is a refactored version of the inference developed for SR0. Featured improvements:

Accommodate to common alea format to take advantage of existing alea functions
Simplify the structure: Instead of having multiple layers in nton.lower, here we only need to add a specific transformation and source to the ces. The rest of the function is handled by existing bases in alea, which users don't need to touch (in nton.lower users need to modify all of the layers in order to add a background or change transformation).
Decouple transformations (smearing, bias, efficiency) from sources. Therefore users can feed their self-defined transformations into the framework without touching the sources.
Enable the fitting for the parameters in transformations: in nton.lower the smearing and energy bias were hard coded and fixed during the fitting. Now they can also be fit if needed.

ces_source.py:

Defines the basic sources needed for 1D CES
Validates the transformation parameters and models
Users can write extended sources if there's a need. But most of time, the basic ones should be enough.

ces_transformation.py:

Defines the basic transformations needed for 1D CES
Users can add transformations/models if there's a need. Those functions may need to be changed frequently because of the update of data/cuts/corrections.

xe133_template.ii.h5

An example template (Hist1d)

unblind_ces_simple.yaml:

An example yaml config including:

A Hist-based component (Xe133)
A Monoenergetic component (test_gaussian)
A Flat component (test_flat)

4_ces_inference.ipynb

An example notebook showing how to build model based on example ymal config and do fittings

*Note: the rest of the changes are just auto formatting because of black . && flake8

…d missing import statement

…l bug in CESTemplateSource class

kdund · 2024-05-02T18:37:02Z

The one

@hammannr to be specific-- you are saying that the alea/simulators.py reuses/duplicates some source .simulate functionality, right? I agree that it is cleanest with the sources handling interpolation etc. to the extent possible.

I think so, yes. I can check again and try to improve this. But if I don't miss anything, I think the simulate method of the source is currently completely circumvented.

The one problem I might see with that is that we want the source classes to be fully fledged bluesky sources also so that part kinda has to stay?

hammannr · 2024-05-02T18:44:41Z

The one

@hammannr to be specific-- you are saying that the alea/simulators.py reuses/duplicates some source .simulate functionality, right? I agree that it is cleanest with the sources handling interpolation etc. to the extent possible.

I think so, yes. I can check again and try to improve this. But if I don't miss anything, I think the simulate method of the source is currently completely circumvented.

The one problem I might see with that is that we want the source classes to be fully fledged bluesky sources also so that part kinda has to stay?

Yes -- my idea is to try to use as much of the source functionality as possible in the simulator. At least right now I don't see the reason (yet) why so much of the functionality of the sources is rewritten in the simulator.

for more information, see https://pre-commit.ci

yuema137 · 2024-05-02T21:32:39Z

Hi @hammannr thanks for the comments! I have done the following tasks:

Add docstrings to the public methods
Revert the formatting commit
Add more explanations to the example notebook

And the following's are ongoing as I need to check if they affect any functions:

Use piecewise interpolation for now
Enable apply_slice_args

For those two points I'm not perfectly sure so further comments would be helpful:

generalized 1D framework: from my view, this is already a generalized framework as the transformations could be overwritten by users. And they could apply any possible transformations to 1d histogram. So the main change would just be renaming the files and classes. Am I missing anything here?
Normalization: The motivation to normalize them into event/year/keV but not event/year/ton/keV is that some of the backgrounds don't scale up with the fiducial mass. In the lower inference, we used to calculate rate multipliers in the inference, which made the code hard to modify. So, I think it's better to separate the statistical inference from the rate multiplier calculation. The rate multiplier should come from external calculations (YAML configs). But if there's a better solution I'm glad to implement

kdund

Hi, @yuema137 awesome to see this PR where we get a dedicated 1D source! Some comments added, I agree with Robert that we should remove specific CES-things to make this useful for any 1D analysis we make :)

kdund · 2024-05-02T17:48:05Z

notebooks/4_ces_inference.ipynb

I think here I'd put some pedagogical comments :)

Also love the 2D likelihood surface :)!

kdund · 2024-05-02T18:52:57Z

alea/ces_source.py

+from multihist import Hist1d
+from alea.ces_transformation import Transformation
+
+MINIMAL_ENERGY_RESOLUTION = 0.05


I do not think this boundary is required-- for fits, it should be set in the config fit boundary, and the bins can be as fine as we'd like and still be valid. (and other 1D fits may have a different "natural" scale for the bins)

kdund · 2024-05-02T19:39:37Z

alea/ces_source.py

+            h = efficiency_transformation.apply_transformation(h)
+        if smearing_transformation is not None:
+            h = smearing_transformation.apply_transformation(h)
+        if bias_transformation is not None:


I think the bias makes more sense to apply in true energy (i.e. before smearing) what do you think?

Actually I think it's better to add in the smeared energy. Because the bias should in principle be caused by some area-dependent effect (like merging SEs and lone hits). So for example if a monoenergetic source produces two S2s, 1.1e4 and 1.2e4 separately, the biases would be slightly different for them even if the true energies are the same.

In general the bias should be a function of true parameters (I agree if we had a measurement of "true quanta released" we could define our bias in terms of that, and the CES could be considered an approximation of that, but it is only an approximation)

kdund · 2024-05-02T21:44:32Z

alea/ces_source.py

+        return h
+
+    def _transform_histogram(self, h: Hist1d):
+        # Only efficiency is applicable for flat source


Disagree-- you could have an efficiency defined in true energy and then a smearing on top, I think?

Similar to bias, efficiency is applied after the smearing. The smearing in our detector is mainly due to the Poisson fluctuation during the production of n_e and n_gamma which is before any detector effects start to get involved. This is the main reason that we set the smearing as the first transformation

IMO the efficiency should be the probability to detect an event given a certain initial recoil energy. This is what we use for WIMPs, and it is a well-defined probability (that does include the effects you list here as important inputs!).
I think attempting to apply efficiencies based on reconstructed variables on the basis that those reconstructed variables approximate (but does not give) the underlying number of quanta that are important for the probability to be detected

Hi Knut, to clarify, the efficiency here is only projected in CES space. For S2 threshold and S1 efficiency, we get the CES projected values during the band fitting. For cut acceptance which is data-driven, the values are naturally only in CES. Defining everything in true energy is possible but the gain is limited. It will make every efficiency study entangled with the band fitting which is not necessary for lower or other studies performed in CES.

kdund · 2024-05-02T21:48:00Z

alea/ces_transformation.py

+        if bins.size <= 1:
+            raise ValueError("bins must have at least 2 elements")
+        bin_centers = 0.5 * (bins[1:] + bins[:-1])
+        data = stats.norm.pdf(


With coarse binning, I would propose instead defining the data as the difference of the CDF evaluated at the upper edge of the bin and evaluated at the lower edge of the bin

for more information, see https://pre-commit.ci

coveralls · 2024-12-08T05:50:23Z

Pull Request Test Coverage Report for Build 12290764765

Warning: This coverage report may be inaccurate.

This pull request's base commit is no longer the HEAD commit of its target branch. This means it includes changes from outside the original pull request, including, potentially, unrelated coverage changes.

For more information on this, see Tracking coverage changes with pull request builds.
To avoid this issue with future PRs, see these Recommended CI Configurations.
For a quick fix, rebase this PR at GitHub. Your next report should be accurate.

Details

0 of 207 (0.0%) changed or added relevant lines in 2 files are covered.
7 unchanged lines in 1 file lost coverage.
Overall coverage decreased (-8.8%) to 80.862%

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
alea/ces_transformation.py	0	63	0.0%
alea/ces_source.py	0	144	0.0%

Files with Coverage Reduction	New Missed Lines	%
alea/models/blueice_extended_model.py	7	96.84%

Totals
Change from base Build 12040815089:	-8.8%
Covered Lines:	1707
Relevant Lines:	2111

💛 - Coveralls

…values

for more information, see https://pre-commit.ci

yuema137 added 30 commits March 13, 2024 22:30

refractors smearing and acceptance functions used in lower

a5e2a2b

Add .venv to .gitignore

21c2226

Add CESTemplateSource class to ces_source.py

02a2b29

Add new YAML configuration files and notebook for CES inference

21cf1eb

Fix optional parameter type in smearing_hist_gaussian function and ad…

71fdc56

…d missing import statement

Merge branch 'main' of github.com:XENONnT/alea into cesinf

6613818

Update biasing_hist_arctan and biasing_hist_sigmoid default parameters

06327e4

Finished the first template source for ces space

cc1931a

Refactor CESTemplateSource class and add new methods

85bc7b5

Add new files and update ces_source.py and ces_functions.py

bd4536e

Add xe133_template.ii.h5 file

c6a634b

Add support for additional transformations and fix parameter retrieva…

40189f3

…l bug in CESTemplateSource class

Fix biasing calculation in CES functions

78dc09e

Add mypy configuration file

786d664

Update livetime description and xe133 rate multiplier value

d7c4917

Add example notebook for the ces inference

6058072

update ces notebook

c9871b8

refactor build_histogram: break into several functions

1c14ad2

delete unused files

55d6b07

Add function to fast smear monoenergy histogram

25ac3b5

Add gaussian source and flat source to the model

3fd09df

Update example notebook showing several different sources

392ef1e

Merge branch 'main' of github.com:XENONnT/alea into cesinf

4cc19c1

change name for the monoenergy source

1cd0178

Move ces funcs to a temp folder

3d671e0

Update ces_source

321fe1f

fix typo

bfd25a0

update ces example yaml

417b455

update example ces peak energy

ebf2ce2

update the example notebook to show the fitted smearing model

4690f23

yuema137 added 2 commits May 2, 2024 13:12

Add docstring to transformations

2889bd9

update example yaml

515c200

yuema137 and others added 7 commits May 2, 2024 14:16

change sequence of transformations

1122ec4

change sequence of transformations

469338b

modify parameter name

322499b

update example ymal

12872e3

update example yaml

bda4f6c

add explanations to the example notebook

38586ef

[pre-commit.ci] auto fixes from pre-commit.com hooks

e449e29

for more information, see https://pre-commit.ci

kdund requested changes May 2, 2024

View reviewed changes

yuema137 and others added 8 commits May 9, 2024 19:52

change histogram type to float

01df110

Merge branch 'cesinf' of github.com:XENONnT/alea into cesinf

6d6907d

[pre-commit.ci] auto fixes from pre-commit.com hooks

5dcbd44

for more information, see https://pre-commit.ci

formatting

286b7ce

Merge branch 'cesinf' of github.com:XENONnT/alea into cesinf

08ba7fa

[pre-commit.ci] auto fixes from pre-commit.com hooks

fcdf7c8

for more information, see https://pre-commit.ci

Merge branch 'main' of github.com:XENONnT/alea into cesinf

d57a780

Merge branch 'main' into cesinf

437786a

yuema137 and others added 8 commits December 11, 2024 22:57

add wrapper to make the lookup function to return 0 for out-of-range …

ce54e48

…values

Merge branch 'cesinf' of github.com:XENONnT/alea into cesinf

02aee11

[pre-commit.ci] auto fixes from pre-commit.com hooks

4e28bc1

for more information, see https://pre-commit.ci

add rebin function to make the pdf histogram consistent

f9aabb8

add binned ces config

684a52d

fix the source_wise_interpolation overriding

72db646

add binned ces fit demo notebook

fbe7750

Merge branch 'cesinf' of github.com:XENONnT/alea into cesinf

72cc0b9

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add sources and example notebook for 1D CES inference #161

Add sources and example notebook for 1D CES inference #161

yuema137 commented May 1, 2024 •

edited

Loading

kdund commented May 2, 2024

hammannr commented May 2, 2024

yuema137 commented May 2, 2024 •

edited

Loading

kdund left a comment

kdund May 2, 2024

kdund May 2, 2024

kdund May 2, 2024

kdund May 2, 2024

yuema137 May 3, 2024 •

edited

Loading

kdund May 7, 2024

kdund May 2, 2024

yuema137 May 3, 2024 •

edited

Loading

kdund May 7, 2024

yuema137 May 7, 2024

kdund May 2, 2024

coveralls commented Dec 8, 2024 •

edited

Loading

Add sources and example notebook for 1D CES inference #161

Are you sure you want to change the base?

Add sources and example notebook for 1D CES inference #161

Conversation

yuema137 commented May 1, 2024 • edited Loading

Note: This PR doesn't affect existing functions in Alea but only adds 1D CES inference.

ces_source.py:

ces_transformation.py:

xe133_template.ii.h5

unblind_ces_simple.yaml:

4_ces_inference.ipynb

kdund commented May 2, 2024

hammannr commented May 2, 2024

yuema137 commented May 2, 2024 • edited Loading

kdund left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

yuema137 May 3, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

yuema137 May 3, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

coveralls commented Dec 8, 2024 • edited Loading

Pull Request Test Coverage Report for Build 12290764765

Warning: This coverage report may be inaccurate.

Details

💛 - Coveralls

yuema137 commented May 1, 2024 •

edited

Loading

yuema137 commented May 2, 2024 •

edited

Loading

yuema137 May 3, 2024 •

edited

Loading

yuema137 May 3, 2024 •

edited

Loading

coveralls commented Dec 8, 2024 •

edited

Loading