DEPRECATED: General theory covmat
Created by: RosalynLP
Replaces #646. Aim is to allow arbitrary general theory covmats to be loaded from file alongside the scale variation covmat.
Merge request reports
Closed by
(Apr 8, 2025 7:19am UTC)
Activity
Created by: RosalynLP
The main points to work on are:
- Get rid of user-specific location for stored theory covmats
- Allow choice of theory covmat type for scalevar cov (e.g. block diag or diag as well as full) - although this does seem a bit redundant at this stage and I'd be happy to get rid of this functionality
- Allow for the scenario where not all covmats are present
-
Introduce another function for
user_covmatto be loaded from file, so the user can specify any covmat outwith the choices of scalevar, top, higher twist or some combination. - Deal with the changing name of the dumped file which depends on dataspecs
requested review from @enocera
requested review from @enocera
requested review from @enocera
Created by: RosalynLP
@Zaharid as I mentioned at the code call I am having problems using the loader to load the theory covmat.
I uploaded a covariance matrix for top mass uncertainty using
validphys, found at https://vp.nnpdf.science/IeGM9CY8RxGcb5r6bIEYlQ==/topthcovmat.csv.I am trying to load this using the loader in
theorycovariance.construction.pyline 456, but get the errorFile "/home/s1303034/nnpdf/validphys2/src/validphys/loader.py", line 520, in check_vp_output_file raise LoadFailedError(f"Could not find '{filename}'") from e validphys.loader.LoadFailedError: Could not find 'IeGM9CY8RxGcb5r6bIEYlQ==/topthcovmat.csv'and the same if I use the full address with https://vp.nnpdf.science in front.
Is this because this isn't strictly an output file?
Created by: wilsonmr
@RosalynLP I can download it fine using the
FallbackLoaderare you just usingLoader? Note that this just checks for the file locallyprobs best practice would be to create a production rule which uses the loader of the config class because that will check locally and remotely
Created by: wilsonmr
EDIT: ignore the fact that I don't load it correctly, I was just checking the file was a csv
>>> l = FallbackLoader() >>> l.check_vp_output_file("IeGM9CY8RxGcb5r6bIEYlQ==/topthcovmat.csv") PosixPath('/Users/michael/conda/envs/nnpdf/share/NNPDF/vp-cache/IeGM9CY8RxGcb5r6bIEYlQ==/topthcovmat.csv') >>> import pandas as pd >>> pd.read_csv('/Users/michael/conda/envs/nnpdf/share/NNPDF/vp-cache/IeGM9CY8RxGcb5r6bIEYlQ==/topthcovmat.csv') experiment ... BIGEXP.25 0 dataset ... CMSTOPDIFF8TEVTTRAPNORM 1 id ... 9 2 experiment ... NaN 3 BIGEXP ... 0.0715461 4 BIGEXP ... 0.103566 5 BIGEXP ... 0.361396 6 BIGEXP ... 0.0715461 7 BIGEXP ... 0.103566 8 BIGEXP ... 0.361396 9 BIGEXP ... 4.76801e-05 10 BIGEXP ... 0.000110914 11 BIGEXP ... 0.000155474 12 BIGEXP ... 0.000189919 13 BIGEXP ... 0.000208396 14 BIGEXP ... 0.000212287 15 BIGEXP ... 0.000189961 16 BIGEXP ... 0.000155095 17 BIGEXP ... 0.000110225 18 BIGEXP ... 4.80463e-05 19 BIGEXP ... 3.63837e-05 20 BIGEXP ... 0.00014501 21 BIGEXP ... 0.000193685 22 BIGEXP ... 0.000241234 23 BIGEXP ... 0.000262788 24 BIGEXP ... 0.000263333 25 BIGEXP ... 0.000240536 26 BIGEXP ... 0.000196785 27 BIGEXP ... 0.000142848 28 BIGEXP ... 3.64091e-05 [29 rows x 29 columns] >>>Created by: wilsonmr
Urm I'm not sure exactly what you mean there but the point is that if I were to use e.g
check_fitthe loader will return aFitSpecbecause it's a well defined object. If we look atcheck_vp_output_fileit returns the path to the vp-output file you requested (which is really the best it can do because this could be a table, a plot, something else). But this isn't the same as a hardcoded local path, and if we use theFallbackLoaderit also downloads the actual file to some sensible place so schematically you could have:from validphys.loader import FallbackLoader l = FallbackLoader() path_to_covmat = l.check_vp_output_file("IeGM9CY8RxGcb5r6bIEYlQ==/topthcovmat.csv") covmat = some_special_tableloader_function_that_gets_index_correct(path_to_covmat)Where you wrap that up is kind of up to you. At the moment
validphys.configdoesn't importvalidphys.tableloaderand neither doesvalidphys.loaderIf I look back to the alpha s stuff in
validphys.paramfitsthen this functionality appears to be wrapped up in theConfigParserso perhaps that would be fine?Maybe make the
tableloaderimport inside the production rule which checks the vp_output and loads the table since it seems quite othogonal to the rest of the module.Created by: RosalynLP
Urm I'm not sure exactly what you mean there but the point is that if I were to use e.g
check_fitthe loader will return aFitSpecbecause it's a well defined object. If we look atcheck_vp_output_fileit returns the path to the vp-output file you requested (which is really the best it can do because this could be a table, a plot, something else). But this isn't the same as a hardcoded local path, and if we use theFallbackLoaderit also downloads the actual file to some sensible place so schematically you could have:from validphys.loader import FallbackLoader l = FallbackLoader() path_to_covmat = l.check_vp_output_file("IeGM9CY8RxGcb5r6bIEYlQ==/topthcovmat.csv") covmat = some_special_tableloader_function_that_gets_index_correct(path_to_covmat)Where you wrap that up is kind of up to you. At the moment
validphys.configdoesn't importvalidphys.tableloaderand neither doesvalidphys.loaderIf I look back to the alpha s stuff in
validphys.paramfitsthen this functionality appears to be wrapped up in theConfigParserso perhaps that would be fine?Maybe make the
tableloaderimport inside the production rule which checks the vp_output and loads the table since it seems > quite othogonal to the rest of the module.OK thank you so much, I don't know why I was confused, I think this makes sense now! I already have some table loading function which takes the file and loads it how I want, so that is all fine, I think I was just unsure of the distinction between
LoaderandFallbackLoader.It turns out one of the reasons I was having problems was my .netrc file was not up to date, so it was unable to find the file because I didn't have the access to the server.
Created by: RosalynLP
Although the covmats can now be loaded from the server they are set in stone as being from the output of a specific vp upload. For a generic "user covmat" the user might want to make an arbitrary theory covmat and include it in a fit. They would then need to upload their own covmat to the server and alter the vp path in the code in order to use it, which is not ideal. Is there any way in which we could keep a permanent link to a covmat folder which could then be added to at a later date, so that people could add their own covmats? Then maybe they could pass in the filename in the runcard and their chosen covmat name could be loaded from this folder?
mentioned in merge request !991 (merged)
