669 lines
36 KiB
Python
669 lines
36 KiB
Python
import csv
|
|
|
|
from sqlalchemy import Engine, select
|
|
from sqlalchemy.orm import Session
|
|
from tqdm import tqdm
|
|
|
|
from nupes.cache import get_file
|
|
from nupes.models import BureauVote
|
|
from nupes.models.legislatives2022 import *
|
|
|
|
|
|
def importer_resultats_bv(engine: Engine, verbose: bool = False) -> None:
|
|
tours = [(1, "https://www.data.gouv.fr/fr/datasets/r/a1f73b85-8194-44f4-a2b7-c343edb47d32"),
|
|
(2, "https://www.data.gouv.fr/fr/datasets/r/96ffddda-59b4-41b8-a6a3-dfe1adb7fa36")]
|
|
|
|
with Session(engine) as session:
|
|
candidats = {(candidat.circonscription_id, candidat.numero): candidat
|
|
for candidat in session.execute(select(CandidatLegislatives2022)).scalars().all()}
|
|
bv_ids = {(bv.commune_code, bv.code_bureau): bv.id
|
|
for bv in session.execute(select(BureauVote)).scalars().all()}
|
|
|
|
for tour, file_url in tours:
|
|
file = get_file("https://www.data.gouv.fr/fr/datasets/r/a1f73b85-8194-44f4-a2b7-c343edb47d32",
|
|
"resultats-legislatives-2022-t1-par-bureau-de-vote.csv")
|
|
|
|
with file.open('r', encoding="ISO-8859-1") as f:
|
|
reader = csv.reader(f, delimiter=';')
|
|
next(reader)
|
|
with Session(engine) as session:
|
|
for row in tqdm(reader, desc=f"Bureau de vote tour {tour}", disable=not verbose):
|
|
dpt_code = row[0]
|
|
|
|
match dpt_code:
|
|
case "ZA":
|
|
dpt_code = "971"
|
|
case "ZB":
|
|
dpt_code = "972"
|
|
case "ZC":
|
|
dpt_code = "973"
|
|
case "ZD":
|
|
dpt_code = "974"
|
|
case "ZS":
|
|
dpt_code = "975"
|
|
case "ZM":
|
|
dpt_code = "976"
|
|
case "ZX":
|
|
dpt_code = "977"
|
|
case "ZW":
|
|
dpt_code = "986"
|
|
case "ZP":
|
|
dpt_code = "987"
|
|
case "ZN":
|
|
dpt_code = "988"
|
|
|
|
if dpt_code == "987" or dpt_code == "988":
|
|
# Les communes de Polynésie française ne sont pas importées
|
|
continue
|
|
|
|
com_code = str(dpt_code + row[4][(1 if len(dpt_code) == 3 else 0):]).zfill(5)
|
|
|
|
if com_code == "98601":
|
|
# 3 royaumes à Wallis-et-Futuna, mais un seul résultat de commune
|
|
com_code = "98611"
|
|
|
|
if com_code.startswith("987"): # Les communes de Polynésie française ne sont pas importées
|
|
continue
|
|
|
|
bv_code = row[6]
|
|
|
|
if (com_code, bv_code) not in bv_ids:
|
|
print(f"Bureau de vote {com_code}_{bv_code} non trouvé")
|
|
continue
|
|
bv_id = bv_ids[(com_code, bv_code)]
|
|
|
|
circo_code = row[2]
|
|
circo_id = f"{dpt_code.zfill(2)}-{circo_code.zfill(2)}"
|
|
|
|
resultats_bv = session.execute(select(ResultatsBureauVoteLegislatives2022)
|
|
.filter_by(bv_id=bv_id)).scalar_one_or_none()
|
|
if not resultats_bv:
|
|
resultats_commune = session.execute(select(ResultatsCommuneLegislatives2022)
|
|
.filter_by(commune_id=com_code)).scalar_one_or_none()
|
|
if not resultats_commune:
|
|
resultats_dpt = session.execute(select(ResultatsDepartementLegislatives2022)
|
|
.filter_by(dpt_id=dpt_code)) \
|
|
.scalar_one_or_none()
|
|
if not resultats_dpt:
|
|
dpt = session.execute(select(Departement).filter_by(code_insee=dpt_code)) \
|
|
.scalar_one()
|
|
resultats_reg = session.execute(
|
|
select(ResultatsRegionLegislatives2022).filter_by(region_id=dpt.region_code)) \
|
|
.scalar_one_or_none()
|
|
if not resultats_reg:
|
|
resultats_france = session.execute(select(ResultatsFranceLegislatives2022)) \
|
|
.scalar_one_or_none()
|
|
if not resultats_france:
|
|
session.add(ResultatsFranceLegislatives2022())
|
|
resultats_france = session.execute(select(ResultatsFranceLegislatives2022)) \
|
|
.scalar_one()
|
|
|
|
resultats_reg = ResultatsRegionLegislatives2022(
|
|
region_id=str(dpt.region_code), resultats_france_id=resultats_france.id)
|
|
session.add(resultats_reg)
|
|
resultats_reg = session.execute(select(ResultatsRegionLegislatives2022)
|
|
.filter_by(region_id=dpt.region_code)).scalar_one()
|
|
|
|
resultats_dpt = ResultatsDepartementLegislatives2022(
|
|
dpt_id=dpt_code, resultats_region_id=resultats_reg.id)
|
|
session.add(resultats_dpt)
|
|
resultats_dpt = session.execute(select(ResultatsDepartementLegislatives2022)
|
|
.filter_by(dpt_id=dpt_code)).scalar_one()
|
|
|
|
resultats_commune = ResultatsCommuneLegislatives2022(commune_id=com_code,
|
|
resultats_dpt_id=resultats_dpt.id)
|
|
session.add(resultats_commune)
|
|
resultats_commune = session.execute(select(ResultatsCommuneLegislatives2022)
|
|
.filter_by(commune_id=com_code)).scalar_one()
|
|
|
|
resultats_circo = session.execute(select(ResultatsCirconscriptionLegislatives2022)
|
|
.filter_by(circo_id=circo_id)).scalar_one_or_none()
|
|
|
|
resultats_bv = ResultatsBureauVoteLegislatives2022(bv_id=bv_id,
|
|
resultats_commune_id=resultats_commune.id,
|
|
resultats_circo_id=resultats_circo.id)
|
|
session.add(resultats_bv)
|
|
resultats_bv = session.execute(select(ResultatsBureauVoteLegislatives2022)
|
|
.filter_by(bv_id=bv_id)).scalar_one()
|
|
|
|
if tour == 1:
|
|
resultats_bv.inscrits_t1 = int(row[7])
|
|
resultats_bv.votants_t1 = int(row[10])
|
|
resultats_bv.abstentions_t1 = int(row[8])
|
|
resultats_bv.exprimes_t1 = int(row[18])
|
|
resultats_bv.blancs_t1 = int(row[12])
|
|
resultats_bv.nuls_t1 = int(row[15])
|
|
elif tour == 2:
|
|
resultats_bv.inscrits_t2 = int(row[7])
|
|
resultats_bv.votants_t2 = int(row[10])
|
|
resultats_bv.abstentions_t2 = int(row[8])
|
|
resultats_bv.exprimes_t2 = int(row[18])
|
|
resultats_bv.blancs_t2 = int(row[12])
|
|
resultats_bv.nuls_t2 = int(row[15])
|
|
|
|
row_voix = row[21:]
|
|
|
|
for i in range(len(row_voix) // 8):
|
|
bloc_voix = row_voix[i * 8:(i + 1) * 8]
|
|
numero = bloc_voix[0]
|
|
voix = int(bloc_voix[5])
|
|
|
|
candidat = candidats[(circo_id, int(numero))]
|
|
|
|
voix_candidat_bv = session.execute(select(VoixBureauVoteLegislatives2022)
|
|
.filter_by(resultats_bureau_vote_id=resultats_bv.id,
|
|
candidat_id=candidat.id)) \
|
|
.scalar_one_or_none()
|
|
if not voix_candidat_bv:
|
|
voix_candidat_bv = VoixBureauVoteLegislatives2022(
|
|
resultats_bureau_vote_id=resultats_bv.id, candidat_id=candidat.id)
|
|
session.add(voix_candidat_bv)
|
|
|
|
if tour == 1:
|
|
voix_candidat_bv.voix_t1 = voix
|
|
elif tour == 2:
|
|
voix_candidat_bv.voix_t2 = voix
|
|
|
|
session.commit()
|
|
|
|
|
|
def importer_resultats_commune(engine: Engine, verbose: bool = False) -> None:
|
|
tours = [(1, "https://www.data.gouv.fr/fr/datasets/r/a9a82bcc-304e-491f-a4f0-c06575113745"),
|
|
(2, "https://www.data.gouv.fr/fr/datasets/r/e79b2e51-0841-4266-b626-57cb271c7a35")]
|
|
|
|
with Session(engine) as session:
|
|
for resultats_commune in session.execute(select(ResultatsCommuneLegislatives2022)).scalars().all():
|
|
resultats_commune.inscrits_t1 = 0
|
|
resultats_commune.votants_t1 = 0
|
|
resultats_commune.abstentions_t1 = 0
|
|
resultats_commune.exprimes_t1 = 0
|
|
resultats_commune.blancs_t1 = 0
|
|
resultats_commune.nuls_t1 = 0
|
|
resultats_commune.inscrits_t2 = 0
|
|
resultats_commune.votants_t2 = 0
|
|
resultats_commune.abstentions_t2 = 0
|
|
resultats_commune.exprimes_t2 = 0
|
|
resultats_commune.blancs_t2 = 0
|
|
resultats_commune.nuls_t2 = 0
|
|
|
|
for voix_nuance in resultats_commune.voix:
|
|
voix_nuance.voix_t1 = 0
|
|
voix_nuance.voix_t2 = 0
|
|
|
|
session.commit()
|
|
|
|
for tour, file_url in tours:
|
|
file = get_file(file_url,
|
|
f"resultats-legislatives-2022-t{tour}-par-commune.csv")
|
|
|
|
with file.open('r', encoding="ISO-8859-1") as f:
|
|
reader = csv.reader(f, delimiter=';')
|
|
next(reader)
|
|
with Session(engine) as session:
|
|
for row in tqdm(reader, desc=f"Commune tour {tour}", disable=not verbose):
|
|
dpt_code = row[0]
|
|
|
|
match dpt_code:
|
|
case "ZA":
|
|
dpt_code = "971"
|
|
case "ZB":
|
|
dpt_code = "972"
|
|
case "ZC":
|
|
dpt_code = "973"
|
|
case "ZD":
|
|
dpt_code = "974"
|
|
case "ZS":
|
|
dpt_code = "975"
|
|
case "ZM":
|
|
dpt_code = "976"
|
|
case "ZX":
|
|
dpt_code = "977"
|
|
case "ZW":
|
|
dpt_code = "986"
|
|
case "ZP":
|
|
dpt_code = "987"
|
|
case "ZN":
|
|
dpt_code = "988"
|
|
|
|
if dpt_code == "987" or dpt_code == "988":
|
|
# Les communes de Polynésie française ne sont pas importées
|
|
continue
|
|
|
|
com_code = str(dpt_code + row[4][(1 if len(dpt_code) == 3 else 0):]).zfill(5)
|
|
|
|
if com_code == "98601":
|
|
# 3 royaumes à Wallis-et-Futuna, mais un seul résultat de commune
|
|
com_code = "98611"
|
|
|
|
resultats_commune = session.execute(select(ResultatsCommuneLegislatives2022)
|
|
.filter_by(commune_id=com_code)).scalar_one_or_none()
|
|
if not resultats_commune:
|
|
resultats_dpt = session.execute(select(ResultatsDepartementLegislatives2022)
|
|
.filter_by(dpt_id=dpt_code)).scalar_one_or_none()
|
|
if not resultats_dpt:
|
|
dpt = session.execute(select(Departement)
|
|
.filter_by(code_insee=f"{dpt_code.zfill(2)}")).scalar_one()
|
|
resultats_reg = session.execute(select(ResultatsRegionLegislatives2022)
|
|
.filter_by(region_id=dpt.region_code)).scalar_one_or_none()
|
|
if not resultats_reg:
|
|
resultats_france = session.execute(select(ResultatsFranceLegislatives2022)) \
|
|
.scalar_one_or_none()
|
|
if not resultats_france:
|
|
session.add(ResultatsFranceLegislatives2022())
|
|
resultats_france = session.execute(select(ResultatsFranceLegislatives2022)) \
|
|
.scalar_one()
|
|
|
|
resultats_reg = ResultatsRegionLegislatives2022(region_id=str(dpt.region_code),
|
|
resultats_france_id=resultats_france.id)
|
|
session.add(resultats_reg)
|
|
resultats_reg = session.execute(select(ResultatsRegionLegislatives2022)
|
|
.filter_by(region_id=dpt.region_code)).scalar_one()
|
|
|
|
resultats_dpt = ResultatsDepartementLegislatives2022(dpt_id=dpt_code,
|
|
resultats_region_id=resultats_reg.id)
|
|
session.add(resultats_dpt)
|
|
resultats_dpt = session.execute(select(ResultatsDepartementLegislatives2022)
|
|
.filter_by(dpt_id=dpt_code)).scalar_one()
|
|
|
|
resultats_commune = ResultatsCommuneLegislatives2022(commune_id=com_code,
|
|
resultats_dpt_id=resultats_dpt.id)
|
|
session.add(resultats_commune)
|
|
resultats_commune = session.execute(select(ResultatsCommuneLegislatives2022)
|
|
.filter_by(commune_id=com_code)).scalar_one()
|
|
|
|
if tour == 1:
|
|
resultats_commune.inscrits_t1 = int(row[7])
|
|
resultats_commune.votants_t1 = int(row[10])
|
|
resultats_commune.abstentions_t1 = int(row[8])
|
|
resultats_commune.exprimes_t1 = int(row[18])
|
|
resultats_commune.blancs_t1 = int(row[12])
|
|
resultats_commune.nuls_t1 = int(row[15])
|
|
elif tour == 2:
|
|
resultats_commune.inscrits_t2 = int(row[7])
|
|
resultats_commune.votants_t2 = int(row[10])
|
|
resultats_commune.abstentions_t2 = int(row[8])
|
|
resultats_commune.exprimes_t2 = int(row[18])
|
|
resultats_commune.blancs_t2 = int(row[12])
|
|
resultats_commune.nuls_t2 = int(row[15])
|
|
|
|
row_voix = row[21:]
|
|
|
|
for i in range(len(row_voix) // 8):
|
|
bloc_voix = row_voix[i * 8:(i + 1) * 8]
|
|
nuance_code = bloc_voix[4]
|
|
voix = int(bloc_voix[5])
|
|
|
|
voix_nuance_commune = session.execute(select(VoixCommuneLegislatives2022)
|
|
.filter_by(resultats_commune_id=resultats_commune.id,
|
|
nuance_id=nuance_code)) \
|
|
.scalar_one_or_none()
|
|
if not voix_nuance_commune:
|
|
voix_nuance_commune = VoixCommuneLegislatives2022(resultats_commune_id=resultats_commune.id,
|
|
nuance_id=nuance_code,
|
|
voix_t1=0, voix_t2=0)
|
|
session.add(voix_nuance_commune)
|
|
|
|
if tour == 1:
|
|
voix_nuance_commune.voix_t1 += voix
|
|
elif tour == 2:
|
|
voix_nuance_commune.voix_t2 += voix
|
|
|
|
session.commit()
|
|
|
|
|
|
def importer_resultats_circo(engine: Engine, verbose: bool = False) -> None:
|
|
tours = [(1, "https://www.data.gouv.fr/fr/datasets/r/114b13e8-7ff9-437f-9ec8-7a29258a80e3"),
|
|
(2, "https://www.data.gouv.fr/fr/datasets/r/436b9eaa-64e1-46bd-ad1d-9638ebbd6caf")]
|
|
|
|
for tour, file_url in tours:
|
|
file = get_file(file_url, f"resultats-legislatives-2022-t{tour}-par-circonscription.csv")
|
|
|
|
with file.open('r', encoding="ISO-8859-1") as f:
|
|
reader = csv.reader(f, delimiter=';')
|
|
next(reader)
|
|
with Session(engine) as session:
|
|
for row in tqdm(reader, desc=f"Circonscription tour {tour}", disable=not verbose):
|
|
dpt_code = row[0]
|
|
circo_code = row[2]
|
|
|
|
match dpt_code:
|
|
case "ZA":
|
|
dpt_code = "971"
|
|
case "ZB":
|
|
dpt_code = "972"
|
|
case "ZC":
|
|
dpt_code = "973"
|
|
case "ZD":
|
|
dpt_code = "974"
|
|
case "ZS":
|
|
dpt_code = "975"
|
|
case "ZM":
|
|
dpt_code = "976"
|
|
case "ZX":
|
|
dpt_code = "977"
|
|
case "ZW":
|
|
dpt_code = "986"
|
|
case "ZP":
|
|
dpt_code = "987"
|
|
case "ZN":
|
|
dpt_code = "988"
|
|
|
|
circo_id = f"{dpt_code.zfill(2)}-{circo_code.zfill(2)}"
|
|
|
|
resultats_circo = session.execute(select(ResultatsCirconscriptionLegislatives2022)
|
|
.filter_by(circo_id=circo_id)).scalar_one_or_none()
|
|
if not resultats_circo:
|
|
resultats_dpt = session.execute(select(ResultatsDepartementLegislatives2022)
|
|
.filter_by(dpt_id=dpt_code)).scalar_one_or_none()
|
|
if not resultats_dpt:
|
|
dpt = session.execute(select(Departement).filter_by(code_insee=f"{dpt_code.zfill(2)}")) \
|
|
.scalar_one()
|
|
resultats_reg = session.execute(select(ResultatsRegionLegislatives2022)
|
|
.filter_by(region_id=dpt.region_code)) \
|
|
.scalar_one_or_none()
|
|
if not resultats_reg:
|
|
resultats_france = session.execute(select(ResultatsFranceLegislatives2022)) \
|
|
.scalar_one_or_none()
|
|
if not resultats_france:
|
|
session.add(ResultatsFranceLegislatives2022())
|
|
resultats_france = session.execute(select(ResultatsFranceLegislatives2022)).scalar_one()
|
|
|
|
resultats_reg = ResultatsRegionLegislatives2022(region_id=str(dpt.region_code),
|
|
resultats_france_id=resultats_france.id)
|
|
session.add(resultats_reg)
|
|
resultats_reg = session.execute(select(ResultatsRegionLegislatives2022)
|
|
.filter_by(region_id=dpt.region_code)).scalar_one()
|
|
|
|
resultats_dpt = ResultatsDepartementLegislatives2022(dpt_id=dpt_code,
|
|
resultats_region_id=resultats_reg.id)
|
|
session.add(resultats_dpt)
|
|
resultats_dpt = session.execute(select(ResultatsDepartementLegislatives2022)
|
|
.filter_by(dpt_id=dpt_code)).scalar_one()
|
|
|
|
resultats_circo = ResultatsCirconscriptionLegislatives2022(
|
|
circo_id=circo_id, resultats_departement_id=resultats_dpt.id)
|
|
session.add(resultats_circo)
|
|
resultats_circo = session.execute(select(ResultatsCirconscriptionLegislatives2022)
|
|
.filter_by(circo_id=circo_id)).scalar_one()
|
|
|
|
if tour == 1:
|
|
resultats_circo.inscrits_t1 = int(row[5])
|
|
resultats_circo.votants_t1 = int(row[8])
|
|
resultats_circo.abstentions_t1 = int(row[6])
|
|
resultats_circo.exprimes_t1 = int(row[16])
|
|
resultats_circo.blancs_t1 = int(row[10])
|
|
resultats_circo.nuls_t1 = int(row[13])
|
|
elif tour == 2:
|
|
resultats_circo.inscrits_t2 = int(row[5])
|
|
resultats_circo.votants_t2 = int(row[8])
|
|
resultats_circo.abstentions_t2 = int(row[6])
|
|
resultats_circo.exprimes_t2 = int(row[16])
|
|
resultats_circo.blancs_t2 = int(row[10])
|
|
resultats_circo.nuls_t2 = int(row[13])
|
|
|
|
row_voix = row[19:]
|
|
|
|
for i in range(len(row_voix) // 9):
|
|
bloc_voix = row_voix[i * 9:(i + 1) * 9]
|
|
numero = bloc_voix[0]
|
|
voix = int(bloc_voix[5])
|
|
|
|
candidat = session.execute(select(CandidatLegislatives2022)
|
|
.filter_by(circonscription_id=circo_id, numero=numero)) \
|
|
.scalar_one_or_none()
|
|
|
|
voix_candidat_circo = session.execute(select(VoixCirconscriptionLegislatives2022)
|
|
.filter_by(resultats_circonscription_id=resultats_circo.id,
|
|
candidat_id=candidat.id)) \
|
|
.scalar_one_or_none()
|
|
if not voix_candidat_circo:
|
|
voix_candidat_circo = VoixCirconscriptionLegislatives2022(
|
|
resultats_circonscription_id=resultats_circo.id, candidat_id=candidat.id)
|
|
session.add(voix_candidat_circo)
|
|
|
|
if tour == 1:
|
|
voix_candidat_circo.voix_t1 = voix
|
|
elif tour == 2:
|
|
voix_candidat_circo.voix_t2 = voix
|
|
|
|
session.commit()
|
|
|
|
|
|
def importer_resultats_departement(engine: Engine, verbose: bool = False) -> None:
|
|
tours = [(1, "https://www.data.gouv.fr/fr/datasets/r/4acba699-e512-4dc5-87c7-ba74db773633"),
|
|
(2, "https://www.data.gouv.fr/fr/datasets/r/bc76bd72-cf65-45b2-b651-7716dc3c5520")]
|
|
|
|
for tour, file_url in tours:
|
|
file = get_file(file_url, f"resultats-legislatives-2022-t{tour}-par-departement.csv")
|
|
|
|
with file.open('r', encoding="ISO-8859-1") as f:
|
|
reader = csv.reader(f, delimiter=';')
|
|
next(reader)
|
|
with Session(engine) as session:
|
|
for row in tqdm(reader, desc=f"Département tour {tour}", disable=not verbose):
|
|
dpt_code = row[0]
|
|
|
|
match dpt_code:
|
|
case "ZA":
|
|
dpt_code = "971"
|
|
case "ZB":
|
|
dpt_code = "972"
|
|
case "ZC":
|
|
dpt_code = "973"
|
|
case "ZD":
|
|
dpt_code = "974"
|
|
case "ZS":
|
|
dpt_code = "975"
|
|
case "ZM":
|
|
dpt_code = "976"
|
|
case "ZX":
|
|
dpt_code = "977"
|
|
case "ZW":
|
|
dpt_code = "986"
|
|
case "ZP":
|
|
dpt_code = "987"
|
|
case "ZN":
|
|
dpt_code = "988"
|
|
|
|
resultats_dpt = session.execute(select(ResultatsDepartementLegislatives2022)
|
|
.filter_by(dpt_id=dpt_code)).scalar_one_or_none()
|
|
if not resultats_dpt:
|
|
dpt = session.execute(select(Departement)
|
|
.filter_by(code_insee=f"{dpt_code.zfill(2)}")) \
|
|
.scalar_one()
|
|
resultats_reg = session.execute(select(ResultatsRegionLegislatives2022)
|
|
.filter_by(region_id=dpt.region_code)) \
|
|
.scalar_one_or_none()
|
|
if not resultats_reg:
|
|
resultats_france = session.execute(select(ResultatsFranceLegislatives2022)) \
|
|
.scalar_one_or_none()
|
|
if not resultats_france:
|
|
session.add(ResultatsFranceLegislatives2022())
|
|
resultats_france = session.execute(select(ResultatsFranceLegislatives2022)).scalar_one()
|
|
|
|
resultats_reg = ResultatsRegionLegislatives2022(region_id=str(dpt.region_code),
|
|
resultats_france_id=resultats_france.id)
|
|
session.add(resultats_reg)
|
|
resultats_reg = session.execute(select(ResultatsRegionLegislatives2022)
|
|
.filter_by(region_id=dpt.region_code)).scalar_one()
|
|
|
|
resultats_dpt = ResultatsDepartementLegislatives2022(dpt_id=dpt_code,
|
|
resultats_region_id=resultats_reg.id)
|
|
session.add(resultats_dpt)
|
|
resultats_dpt = session.execute(select(ResultatsDepartementLegislatives2022)
|
|
.filter_by(dpt_id=dpt_code)).scalar_one()
|
|
|
|
if tour == 1:
|
|
resultats_dpt.inscrits_t1 = int(row[3])
|
|
resultats_dpt.votants_t1 = int(row[6])
|
|
resultats_dpt.abstentions_t1 = int(row[4])
|
|
resultats_dpt.exprimes_t1 = int(row[14])
|
|
resultats_dpt.blancs_t1 = int(row[8])
|
|
resultats_dpt.nuls_t1 = int(row[11])
|
|
elif tour == 2:
|
|
resultats_dpt.inscrits_t2 = int(row[3])
|
|
resultats_dpt.votants_t2 = int(row[6])
|
|
resultats_dpt.abstentions_t2 = int(row[4])
|
|
resultats_dpt.exprimes_t2 = int(row[14])
|
|
resultats_dpt.blancs_t2 = int(row[8])
|
|
resultats_dpt.nuls_t2 = int(row[11])
|
|
|
|
row_voix = row[17:]
|
|
|
|
for i in range(len(row_voix) // 5):
|
|
bloc_voix = row_voix[i * 5:(i + 1) * 5]
|
|
nuance_code = bloc_voix[0]
|
|
voix = int(bloc_voix[1])
|
|
|
|
voix_nuance_dpt = session.execute(select(VoixDepartementLegislatives2022)
|
|
.filter_by(resultats_departement_id=resultats_dpt.id,
|
|
nuance_id=nuance_code)) \
|
|
.scalar_one_or_none()
|
|
if not voix_nuance_dpt:
|
|
voix_nuance_dpt = VoixDepartementLegislatives2022(resultats_departement_id=resultats_dpt.id,
|
|
nuance_id=nuance_code)
|
|
session.add(voix_nuance_dpt)
|
|
|
|
if tour == 1:
|
|
voix_nuance_dpt.voix_t1 = voix
|
|
elif tour == 2:
|
|
voix_nuance_dpt.voix_t2 = voix
|
|
|
|
session.commit()
|
|
|
|
|
|
def importer_resultats_region(engine: Engine, verbose: bool = False) -> None:
|
|
tours = [(1, "https://www.data.gouv.fr/fr/datasets/r/f79576c9-9ddc-43f9-8d99-f0856550054e"),
|
|
(2, "https://www.data.gouv.fr/fr/datasets/r/fdba421a-d4e9-4f38-9753-982a7c065911")]
|
|
|
|
for tour, file_url in tours:
|
|
file = get_file(file_url, f"resultats-legislatives-2022-t{tour}-par-region.csv")
|
|
|
|
with file.open('r', encoding="ISO-8859-1") as f:
|
|
reader = csv.reader(f, delimiter=';')
|
|
next(reader)
|
|
|
|
with Session(engine) as session:
|
|
for row in tqdm(reader, desc=f"Région tour {tour}", disable=not verbose):
|
|
reg_code = row[0]
|
|
|
|
resultats_reg = session.execute(select(ResultatsRegionLegislatives2022)
|
|
.filter_by(region_id=reg_code)).scalar_one_or_none()
|
|
if not resultats_reg:
|
|
resultats_france = session.execute(select(ResultatsFranceLegislatives2022)).scalar_one_or_none()
|
|
if not resultats_france:
|
|
session.add(ResultatsFranceLegislatives2022())
|
|
resultats_france = session.execute(select(ResultatsFranceLegislatives2022)).scalar_one()
|
|
|
|
resultats_reg = ResultatsRegionLegislatives2022(region_id=reg_code,
|
|
resultats_france_id=resultats_france.id)
|
|
session.add(resultats_reg)
|
|
session.commit()
|
|
resultats_reg = session.execute(select(ResultatsRegionLegislatives2022)
|
|
.filter_by(region_id=reg_code)).scalar_one()
|
|
|
|
if tour == 1:
|
|
resultats_reg.inscrits_t1 = int(row[3])
|
|
resultats_reg.votants_t1 = int(row[6])
|
|
resultats_reg.abstentions_t1 = int(row[4])
|
|
resultats_reg.exprimes_t1 = int(row[14])
|
|
resultats_reg.blancs_t1 = int(row[8])
|
|
resultats_reg.nuls_t1 = int(row[11])
|
|
elif tour == 2:
|
|
resultats_reg.inscrits_t2 = int(row[3])
|
|
resultats_reg.votants_t2 = int(row[6])
|
|
resultats_reg.abstentions_t2 = int(row[4])
|
|
resultats_reg.exprimes_t2 = int(row[14])
|
|
resultats_reg.blancs_t2 = int(row[8])
|
|
resultats_reg.nuls_t2 = int(row[11])
|
|
|
|
row_voix = row[17:]
|
|
|
|
for i in range(len(row_voix) // 5):
|
|
bloc_voix = row_voix[i * 5:(i + 1) * 5]
|
|
nuance_code = bloc_voix[0]
|
|
voix = int(bloc_voix[1])
|
|
|
|
voix_nuance_reg = session.execute(select(VoixRegionLegislatives2022)
|
|
.filter_by(resultats_region_id=resultats_reg.id,
|
|
nuance_id=nuance_code)) \
|
|
.scalar_one_or_none()
|
|
if not voix_nuance_reg:
|
|
voix_nuance_reg = VoixRegionLegislatives2022(resultats_region_id=resultats_reg.id,
|
|
nuance_id=nuance_code)
|
|
session.add(voix_nuance_reg)
|
|
|
|
if tour == 1:
|
|
voix_nuance_reg.voix_t1 = voix
|
|
elif tour == 2:
|
|
voix_nuance_reg.voix_t2 = voix
|
|
|
|
session.commit()
|
|
|
|
|
|
def importer_resultats_france(engine: Engine, verbose: bool = False) -> None:
|
|
tours = [(1, "https://www.data.gouv.fr/fr/datasets/r/cbe72f2c-2c56-4251-89cb-213de8bfbea0"),
|
|
(2, "https://www.data.gouv.fr/fr/datasets/r/6639bb39-13ec-49dd-ab4d-1f8227b86c01")]
|
|
for tour, file_url in tours:
|
|
file = get_file(file_url, f"resultats-legislatives-2022-t{tour}-france-entiere.csv")
|
|
|
|
with file.open('r', encoding="ISO-8859-1") as f:
|
|
reader = csv.reader(f, delimiter=';')
|
|
next(reader)
|
|
|
|
with Session(engine) as session:
|
|
for row in tqdm(reader, desc=f"France Entière tour {tour}", disable=not verbose):
|
|
resultats_france = session.execute(select(ResultatsFranceLegislatives2022)).scalar_one_or_none()
|
|
|
|
if not resultats_france:
|
|
session.add(ResultatsFranceLegislatives2022())
|
|
session.commit()
|
|
resultats_france = session.execute(select(ResultatsFranceLegislatives2022)).scalar_one()
|
|
|
|
if tour == 1:
|
|
resultats_france.inscrits_t1 = int(row[3])
|
|
resultats_france.votants_t1 = int(row[6])
|
|
resultats_france.abstentions_t1 = int(row[4])
|
|
resultats_france.exprimes_t1 = int(row[14])
|
|
resultats_france.blancs_t1 = int(row[8])
|
|
resultats_france.nuls_t1 = int(row[11])
|
|
elif tour == 2:
|
|
resultats_france.inscrits_t2 = int(row[3])
|
|
resultats_france.votants_t2 = int(row[6])
|
|
resultats_france.abstentions_t2 = int(row[4])
|
|
resultats_france.exprimes_t2 = int(row[14])
|
|
resultats_france.blancs_t2 = int(row[8])
|
|
resultats_france.nuls_t2 = int(row[11])
|
|
|
|
row_voix = row[17:]
|
|
|
|
for i in range(len(row_voix) // 5):
|
|
bloc_voix = row_voix[i * 5:(i + 1) * 5]
|
|
nuance_code = bloc_voix[0]
|
|
voix = int(bloc_voix[1])
|
|
|
|
voix_nuance_france = session.execute(select(VoixFranceLegislatives2022)
|
|
.filter_by(resultats_france_id=resultats_france.id,
|
|
nuance_id=nuance_code)) \
|
|
.scalar_one_or_none()
|
|
if not voix_nuance_france:
|
|
voix_nuance_france = VoixFranceLegislatives2022(resultats_france_id=resultats_france.id,
|
|
nuance_id=nuance_code)
|
|
session.add(voix_nuance_france)
|
|
|
|
if tour == 1:
|
|
voix_nuance_france.voix_t1 = voix
|
|
elif tour == 2:
|
|
voix_nuance_france.voix_t2 = voix
|
|
|
|
session.commit()
|
|
|
|
|
|
def run(engine: Engine, verbose: bool = False) -> None:
|
|
importer_resultats_france(engine, verbose)
|
|
importer_resultats_region(engine, verbose)
|
|
importer_resultats_departement(engine, verbose)
|
|
importer_resultats_circo(engine, verbose)
|
|
importer_resultats_commune(engine, verbose)
|
|
importer_resultats_bv(engine, verbose)
|