Use NNPDF as a dependency

[ElieHammou]

We want to move away from the NNPDF fork that SIMUnet relies upon currently to use the current NNPDF version as a dependency. We would only modify/add the functions which are specific to SIMUnet. We need to identify SIMUnet-specific functions and either:

• Add them if they are SIMUnet specific function (SIMUnet label), or
• Replace/merge them carefully if they are shared with NNPDF (NNPDF label).

---

From validphys:

• validphys/simunet_analysis.py
  Contains SMEFT post-fit analysis and plotting tools.
  This file should probably be kept as is.

• validphys/bsmnames.py
  Builds names of SMEFT K-factors from runcard info.
  This file should probably be kept as is.

• level0_commondata_wc from validphys/pseudodata.py
  Generates level 0 pseudodata and adds SMEFT K-factors to SM predictions.
  We need to merge our modifications with the current NNPDF version if it has changed.

---

Functions in config.py:

• parse_dataset_input NNPDF
• produce_dataset NNPDF
• produce_fitinputcontext NNPDF
• produce_bsm_fac_initialisations SIMUnet
• produce_bsm_names_to_latex SIMUnet
• produce_bsm_names_to_plot_scales SIMUnet
• produce_contamination_data SIMUnet
• produce_n_simu_parameters SIMUnet
• produce_simu_parameters_linear_combinations SIMUnet
• produce_simu_parameters_names SIMUnet
• produce_simu_parameters_scales SIMUnet

---

Functions in core.py:

• From class DataSetInput:

  • __init__ NNPDF

• From class DataSetSpec:

  • __init__ NNPDF
  • load NNPDF

• From class FitSpec:

  • as_input NNPDF

---

Functions in loader.py:

• From class Loader:
  • check_dataset NNPDF
  • check_fktable NNPDF
  • get_simu_parameters_name_dict SIMUnet

---

Functions in fitdata.py:

• _check_has_bsm_facs SIMUnet
• read_bsm_facs SIMUnet
• Global constant: BSM_FAC_FILE = 'bsm_fac.csv' SIMUnet
• Global variable: pdfs_fits_read_read_bsm_facs = collect(...) SIMUnet

---

Functions in n3fit_data_utils.py:

• common_data_reader_dataset SIMUnet
• parse_simu_parameters_names_CF SIMUnet

---

Functions in n3fit_data.py:

• fitting_data_dict SIMUnet

---

Functions in results.py:

• From class ThPredictionsResult:

  • __init__ NNPDF
  • from_convolution NNPDF

• Top-level functions:

  • __init__ NNPDF
  • dataset_inputs_results NNPDF
  • from_convolution NNPDF
  • one_or_more_results NNPDF
  • pdf_results NNPDF
  • results NNPDF
  • dataset_bsm_factor SIMUnet
  • dataset_inputs_bsm_factor SIMUnet

---

Functions in vp_comparefits.py:

• add_positional_arguments NNPDF
• complete_mapping NNPDF
• get_commandline_arguments NNPDF

---

From n3fit:

• validphys/CombineCfac.py
  Creates the combination layer of SIMUnet

Functions in model_gen.py :

• _generate_experimental_layer NNPDF
  This function builds the extra SMEFT layer in the neural network

---

Functions in model_trainer.py:

• __init__ NNPDF
• _generate_observables NNPDF
• hyperparametrizable NNPDF

---

Functions in perform_fit.py:

• performfit NNPDF

[scarlehoff]

Starting goal, core.py, loader.py, config.py -> get something from which you can do check_dataset API.dataset and get the simunet versions.

1. class SimuNetLoader(Loader) -> def _check_theory_old_or_new(self ... ) (return super().<name of method>)

that function gives you from a) FkTables b) Commondata -> fkspec

that function will call check_fk_from_theory_metadata and check_cfac to produce fktables and cfactors, so it should be enough with overriding def check_fk_from_theory_metadata.

Note: in simunet you were using length(cfactor array) == length(data), 4 fktables -> 4 cfactors

"tests":
a) -> check_dataset() -> should work ootb
b) -> check_fktable() -> more complicated, you should see a FkTableSpec that knows about your simunet cfactors already https://github.com/NNPDF/nnpdf/blob/fef95e6d490656c5c789efb8dcc0f3b1fc0088ff/validphys2/src/validphys/core.py#L570

---

Monkey patching https://github.com/NNPDF/nnpdf/blob/fef95e6d490656c5c789efb8dcc0f3b1fc0088ff/validphys2/examples/API_extension_Pineappl.ipynb