Calage sur la strucure d'age

openfisca_france_data/assets/aggregats/demographie/README transition depuis INSEE.txt

@@ -0,0 +1,2 @@
+Le tableur est issu de https://www.insee.fr/fr/statistiques/5894083?sommaire=5760764 , onglet population. 
+Quelques changements ont été faits pour mieux le parser : conversion en csv, séparateur des milliers mis à "." plutôt qu'à " " (dans Fichier/Options/Options avancées/Utiliser les séparateurs systèmes/Séparateur des milliers), 105 + changé en 105. 

openfisca_france_data/erfs_fpr/scenario.py

@@ -9,6 +9,7 @@ class ErfsFprSurveyScenario(AbstractErfsSurveyScenario):
     # Les variables OpenFisca qu'on va utiliser avec les données en entrée.
     used_as_input_variables = [
         "activite",
+        "age",
         #"autonomie_financiere",
         "categorie_salarie",
         "categorie_non_salarie",

openfisca_france_data/model/calage.py

@@ -1,9 +1,12 @@
-from itertools import izip
+#from itertools import izip
 
 from numpy import arange, array, floor, where
+from numpy import minimum as min_
+import pandas as pd
+from pathlib import Path
 
 from openfisca_france_data.model.base import *  # noqa
-
+from openfisca_france_data import openfisca_france_data_location
 
 class nbinde(Variable):
     value_type = int
@@ -227,3 +230,43 @@ def formula(menage, period, parameters):
         # ratio = (( (typmen15!=res)).sum())/((typmen15!=0).sum())
         # ratio  2.7 % d'erreurs enfant non nés et erreur d'enfants
         return res
+
+def create_dic_calage(year, base_year, liste_variable, **kwargs):
+    dico_calage = dict()
+    assert all(variable in ["age"] for variable in liste_variable), "A variable is asked to be calibrated, but no matching aggregate is implemented"
+    if "age" in liste_variable:
+        dico_calage = create_dic_age(year, base_year, dico_calage, **kwargs)
+    return dico_calage
+
+def create_dic_age(year, base_year, dico_calage, base_initiale = None):
+    demographie_file = Path(
+                openfisca_france_data_location,
+                "openfisca_france_data",
+                "assets",
+                "aggregats",
+                "demographie",
+                "demographie_insee.csv"
+                )
+    df_demographie = pd.read_csv(demographie_file, skipfooter = 9, skiprows = 1, encoding_errors = 'replace', sep = ';', thousands = ".")
+    df_demographie["age_calage"] = df_demographie[df_demographie.columns[0]].astype(int) # On passe par columns à cause du Â
+    for annee in range(base_year, year + 1): # On limite à 100 l'age max, pour éviter l'absence de données
+        df_demographie.loc[df_demographie["age_calage"] == 100, str(annee)] = sum(df_demographie[df_demographie["age_calage"] >= 100][str(annee)])
+    df_demographie = df_demographie[df_demographie["age_calage"] <= 100]
+    if base_initiale is None:
+        dic_age = pd.Series(df_demographie[str(year)].values, index = df_demographie["age_calage"]).to_dict()
+        dico_calage["age_calage"] = dic_age
+        return dico_calage
+
+    df_demographie["variation_totale"] = df_demographie[str(year)] / df_demographie[str(base_year)]
+
+    base_initiale["age_calage"] = base_initiale["age_calage"].astype(int)
+    base_initiale.loc[base_initiale["age_calage"] == 100, "N"] = sum(base_initiale[base_initiale["age_calage"] >= 100]["N"])
+    base_initiale = base_initiale[base_initiale["age_calage"] <= 100]
+
+    cible_finale = pd.merge(df_demographie, base_initiale, on = ("age_calage"), how = "inner")
+    cible_finale["cible"] = round(cible_finale["N"] * cible_finale["variation_totale"])
+    dic_age = pd.Series(cible_finale["cible"].values, index = cible_finale["age_calage"]).to_dict()
+    dico_calage["age_calage"] = dic_age
+    return dico_calage
+
+

openfisca_france_data/reforms/variables_calibration.py

@@ -0,0 +1,31 @@
+import numpy as np
+from numpy import minimum as min_
+
+from openfisca_france.model.base import *  # noqa analysis:ignore
+from openfisca_core.reforms import Reform
+
+import logging
+
+log = logging.getLogger(__name__)
+def create_calibration_tax_benefit_system(survey_scenario = None):
+    class variables_pour_calibration(Reform):
+        name = 'variables_pour_calibration'
+
+        def apply(self):
+
+            class age_calage(Variable):
+                value_type = str
+                entity = Individu
+                label = "Age censuré pour calage"
+                definition_period = YEAR
+                unit = 'years'
+                is_period_size_independent = True
+                set_input = set_input_dispatch_by_period
+
+                def formula(individu, period):
+                    age = individu('age', period.first_month)
+                    return min_(age, 100)
+
+            self.add_variable(age_calage)
+    scenario_for_calibration = variables_pour_calibration(survey_scenario)
+    return scenario_for_calibration

tests/test_calage.py

@@ -0,0 +1,47 @@
+import os
+
+import pytest
+
+from openfisca_survey_manager.calibration import Calibration  # type: ignore
+from openfisca_france_data import openfisca_france_data_location
+from openfisca_france_data.model.calage import create_dic_calage
+from openfisca_france_data.reforms.variables_calibration import create_calibration_tax_benefit_system
+from openfisca_france_data.erfs_fpr.scenario import (  # type: ignore
+    ErfsFprSurveyScenario,
+    )
+from openfisca_france_data import france_data_tax_benefit_system
+
+@pytest.fixture
+def location() -> str:
+    return openfisca_france_data_location
+
+
+def test_calage(survey_scenario, fake_input_data, location, year: int = 2015):
+    # On ititialise le survey scenario
+    survey_scenario2 = ErfsFprSurveyScenario.create(
+    tax_benefit_system = create_calibration_tax_benefit_system(france_data_tax_benefit_system),
+    period = year,
+    )
+    survey_scenario = survey_scenario2
+
+    # On charge les données
+    input_data = fake_input_data(year)
+
+    # On fait la calibration
+    parameters = dict(
+        method = "logit",
+        invlo = 3,
+        up = 3,
+        )
+
+    base_year = 2013
+
+    target_margins_by_variable = create_dic_calage(year, base_year, ["age"])
+
+    calibration_kwargs = {'target_margins_by_variable': target_margins_by_variable, 'parameters': parameters}
+
+    # On initialise le survey scenario
+    survey_scenario.init_from_data(data = dict(input_data_frame = input_data), calibration_kwargs = calibration_kwargs)
+    weight_base = sum(survey_scenario.calculate_variable("wprm", period = base_year))
+
+    assert weight_base < sum(survey_scenario.calculate_variable("wprm", period = year))