-
Notifications
You must be signed in to change notification settings - Fork 4
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
[dbt]:
br_inep_ideb.{uf,brasil}
(#735)
- Loading branch information
Showing
3 changed files
with
276 additions
and
0 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,274 @@ | ||
import pandas as pd | ||
import numpy as np | ||
import os | ||
import zipfile | ||
import basedosdados as bd | ||
|
||
ROOT = os.path.join("models", "br_inep_ideb") | ||
INPUT = os.path.join(ROOT, "input") | ||
TMP = os.path.join(ROOT, "tmp") | ||
OUTPUT = os.path.join(ROOT, "output") | ||
|
||
os.makedirs(INPUT, exist_ok=True) | ||
os.makedirs(TMP, exist_ok=True) | ||
os.makedirs(OUTPUT, exist_ok=True) | ||
|
||
URLS = { | ||
"brasil": "https://download.inep.gov.br/ideb/resultados/divulgacao_brasil_ideb_2023.zip", | ||
"regioes_estados": "https://download.inep.gov.br/ideb/resultados/divulgacao_regioes_ufs_ideb_2023.zip", | ||
# "municipio_anos_iniciais": "https://download.inep.gov.br/educacao_basica/portal_ideb/planilhas_para_download/2021/divulgacao_anos_iniciais_municipios_2021.zip", | ||
# "municipio_anos_finais": "https://download.inep.gov.br/educacao_basica/portal_ideb/planilhas_para_download/2021/divulgacao_anos_finais_municipios_2021.zip", | ||
# "municipio_em": "https://download.inep.gov.br/educacao_basica/portal_ideb/planilhas_para_download/2021/divulgacao_ensino_medio_municipios_2021.zip", | ||
# "escola_anos_iniciais": "https://download.inep.gov.br/educacao_basica/portal_ideb/planilhas_para_download/2021/divulgacao_anos_iniciais_escolas_2021.zip", | ||
# "escola_anos_finais": "https://download.inep.gov.br/educacao_basica/portal_ideb/planilhas_para_download/2021/divulgacao_anos_finais_escolas_2021.zip", | ||
# "escola_em": "https://download.inep.gov.br/educacao_basica/portal_ideb/planilhas_para_download/2021/divulgacao_ensino_medio_escolas_2021.zip", | ||
} | ||
|
||
for _, url in URLS.items(): | ||
os.system(f"cd {INPUT}; curl -O -k {url}") | ||
|
||
|
||
for key, url in URLS.items(): | ||
file = url.split("/")[-1] | ||
file_path = os.path.join(INPUT, file) | ||
output_dir = os.path.join(TMP, key) | ||
|
||
os.makedirs(output_dir, exist_ok=True) | ||
|
||
with zipfile.ZipFile(file_path, "r") as zip_ref: | ||
zip_ref.extractall(output_dir) | ||
|
||
XLSX_BR = os.path.join( | ||
TMP, "brasil", "divulgacao_brasil_ideb_2023", "divulgacao_brasil_ideb_2023.xlsx" | ||
) | ||
|
||
sheet_names_br: list[str] = pd.ExcelFile(XLSX_BR).sheet_names | ||
|
||
COMMON_RENAMES = { | ||
"rede": "rede", | ||
"VL_NOTA_MATEMATICA_2023": "nota_saeb_matematica", | ||
"VL_NOTA_PORTUGUES_2023": "nota_saeb_lingua_portuguesa", | ||
"VL_NOTA_MEDIA_2023": "nota_saeb_media_padronizada", | ||
"VL_INDICADOR_REND_2023": "indicador_rendimento", | ||
"VL_APROVACAO_2023_SI_4": "taxa_aprovacao", | ||
"VL_OBSERVADO_2023": "ideb", | ||
} | ||
|
||
df_brasil = pd.concat( | ||
[ | ||
pd.read_excel(XLSX_BR, sheet_name=sheet_name, skiprows=9)[ | ||
list(COMMON_RENAMES.keys()) | ||
] | ||
.rename(columns=COMMON_RENAMES, errors="raise") # type: ignore | ||
.assign(anos_escolares=sheet_name) | ||
for sheet_name in sheet_names_br | ||
] | ||
) | ||
|
||
df_brasil = df_brasil.pipe(lambda d: d.loc[d["rede"].notna()]).assign(ano=2023) | ||
|
||
df_brasil["rede"].unique() | ||
|
||
df_brasil["rede"] = ( | ||
df_brasil["rede"] | ||
.str.lower() | ||
.replace({"privada (1)": "privada", "pública": "publica"}) | ||
) | ||
|
||
df_brasil["anos_escolares"] = df_brasil["anos_escolares"].replace( | ||
{ | ||
"Brasil (Anos Iniciais)": "iniciais (1-5)", | ||
"Brasil (Anos Finais)": "finais (6-9)", | ||
"Brasil (EM)": "todos (1-4)", | ||
} | ||
) | ||
|
||
df_brasil["projecao"] = np.nan | ||
|
||
df_brasil["ensino"] = df_brasil["anos_escolares"].apply( | ||
lambda v: "medio" if v == "todos (1-4)" else "fundamental" | ||
) | ||
|
||
df_brasil_upstream = bd.read_sql( | ||
"select * from `basedosdados.br_inep_ideb.brasil`", | ||
billing_project_id="basedosdados-dev", | ||
) | ||
|
||
tb_brasil = bd.Table(dataset_id="br_inep_ideb", table_id="brasil") | ||
|
||
tb_brasil_cols_from_bq = tb_brasil._get_columns_from_bq() | ||
|
||
assert len(tb_brasil_cols_from_bq["partition_columns"]) == 0 | ||
|
||
tb_brasil_order_cols: list[str] = [i["name"] for i in tb_brasil_cols_from_bq["columns"]] | ||
|
||
|
||
df_brasil_updated = pd.concat([df_brasil[tb_brasil_order_cols], df_brasil_upstream]) # type: ignore | ||
|
||
OUTPUT_BR = os.path.join(OUTPUT, "brasil.csv") | ||
|
||
df_brasil_updated.to_csv(OUTPUT_BR, index=False) | ||
|
||
tb_brasil.create(OUTPUT_BR, if_table_exists="replace", if_storage_data_exists="replace") | ||
|
||
# Regioes, UFs | ||
|
||
XLSX_REGIOES_UFS = os.path.join( | ||
TMP, | ||
"regioes_estados", | ||
"divulgacao_regioes_ufs_ideb_2023", | ||
"divulgacao_regioes_ufs_ideb_2023.xlsx", | ||
) | ||
|
||
sheet_names_regioes_ufs: list[str] = pd.ExcelFile(XLSX_REGIOES_UFS).sheet_names | ||
|
||
|
||
df_regioes_ufs_latest = pd.concat( | ||
[ | ||
pd.read_excel(XLSX_REGIOES_UFS, sheet_name=sheet_name, skiprows=9)[ | ||
[ | ||
*["Unnamed: 0", "Unnamed: 1"], | ||
*[i for i in COMMON_RENAMES.keys() if i != "rede"], | ||
] | ||
] # type: ignore | ||
.rename(columns={"Unnamed: 0": "uf_regiao", "Unnamed: 1": "rede"}) | ||
.rename(columns=COMMON_RENAMES, errors="raise") | ||
.assign(anos_escolares=sheet_name) | ||
for sheet_name in sheet_names_regioes_ufs | ||
] | ||
) | ||
|
||
df_regioes_ufs_latest = df_regioes_ufs_latest.pipe( | ||
lambda d: d.loc[d["rede"].notna()] | ||
).assign(ano=2023) | ||
|
||
df_regioes_ufs_latest["rede"].unique() | ||
|
||
df_regioes_ufs_latest["rede"] = ( | ||
df_regioes_ufs_latest["rede"] | ||
.str.lower() | ||
.replace( | ||
{ | ||
"privada (1)": "privada", | ||
"privada (2)": "privada", | ||
"total (3)(4)": "total", | ||
"total (4)": "total", | ||
"pública": "publica", | ||
"pública (4)": "publica", | ||
} | ||
) | ||
) | ||
|
||
df_regioes_ufs_latest["anos_escolares"].unique() | ||
|
||
df_regioes_ufs_latest["anos_escolares"] = df_regioes_ufs_latest[ | ||
"anos_escolares" | ||
].replace( | ||
{ | ||
"UF e Regiões (AI)": "iniciais (1-5)", | ||
"UF e Regiões (AF)": "finais (6-9)", | ||
"UF e Regiões (EM)": "todos (1-4)", | ||
} | ||
) | ||
|
||
df_regioes_ufs_latest["projecao"] = np.nan | ||
|
||
df_regioes_ufs_latest["ensino"] = df_regioes_ufs_latest["anos_escolares"].apply( | ||
lambda v: "medio" if v == "todos (1-4)" else "fundamental" | ||
) | ||
|
||
df_regioes_ufs_latest["uf_regiao"].unique() | ||
|
||
SIGLA_UFS_REPLACES = { | ||
"R. G. do Norte": "Rio Grande do Norte", | ||
"R. G. do Sul": "Rio Grande do Sul", | ||
"M. G. do Sul": "Mato Grosso do Sul", | ||
} | ||
|
||
df_regioes_ufs_latest["uf_regiao"] = df_regioes_ufs_latest["uf_regiao"].replace( | ||
SIGLA_UFS_REPLACES | ||
) | ||
|
||
br_dirs = bd.read_sql( | ||
"SELECT * from `basedosdados.br_bd_diretorios_brasil.uf`", | ||
billing_project_id="basedosdados-dev", | ||
) | ||
|
||
assert isinstance(br_dirs, pd.DataFrame) | ||
|
||
## Região | ||
|
||
df_regioes_latest = df_regioes_ufs_latest.loc[ | ||
df_regioes_ufs_latest["uf_regiao"].isin(br_dirs["regiao"].unique().tolist()) | ||
] | ||
|
||
assert len(df_regioes_latest["uf_regiao"].unique()) == 5 | ||
|
||
df_regioes_latest = df_regioes_latest.rename( | ||
columns={"uf_regiao": "regiao"}, errors="raise" | ||
) | ||
|
||
tb_regiao = bd.Table(dataset_id="br_inep_ideb", table_id="regiao") | ||
|
||
tb_regiao_cols_from_bq = tb_regiao._get_columns_from_bq() | ||
|
||
assert len(tb_regiao_cols_from_bq["partition_columns"]) == 0 | ||
|
||
tb_regiao_order_cols: list[str] = [i["name"] for i in tb_regiao_cols_from_bq["columns"]] | ||
|
||
df_regiao_upstream = bd.read_sql( | ||
"select * from `basedosdados.br_inep_ideb.regiao`", | ||
billing_project_id="basedosdados-dev", | ||
) | ||
|
||
df_regiao_updated = pd.concat( | ||
[df_regioes_latest[tb_regiao_order_cols], df_regiao_upstream] # type: ignore | ||
) | ||
|
||
OUTPUT_REGIAO = os.path.join(OUTPUT, "regiao.csv") | ||
|
||
df_regiao_updated.to_csv(OUTPUT_REGIAO, index=False) | ||
|
||
tb_regiao.create( | ||
OUTPUT_REGIAO, if_table_exists="replace", if_storage_data_exists="replace" | ||
) | ||
|
||
## UFs | ||
|
||
df_ufs_latest = df_regioes_ufs_latest.loc[ | ||
df_regioes_ufs_latest["uf_regiao"].isin(br_dirs["nome"].unique().tolist()) | ||
] | ||
|
||
assert len(df_ufs_latest["uf_regiao"].unique()) == 27 | ||
|
||
df_ufs_latest["uf_regiao"].unique() | ||
|
||
|
||
df_ufs_latest = df_ufs_latest.rename(columns={"uf_regiao": "sigla_uf"}, errors="raise") | ||
|
||
df_ufs_latest["sigla_uf"] = df_ufs_latest["sigla_uf"].replace( | ||
{i["nome"]: i["sigla"] for i in br_dirs.to_dict("records")} | ||
) | ||
|
||
tb_uf = bd.Table(dataset_id="br_inep_ideb", table_id="uf") | ||
|
||
tb_uf_cols_from_bq = tb_uf._get_columns_from_bq() | ||
|
||
assert len(tb_uf_cols_from_bq["partition_columns"]) == 0 | ||
|
||
tb_uf_order_cols: list[str] = [i["name"] for i in tb_uf_cols_from_bq["columns"]] | ||
|
||
df_uf_upstream = bd.read_sql( | ||
"select * from `basedosdados.br_inep_ideb.uf`", | ||
billing_project_id="basedosdados-dev", | ||
) | ||
|
||
df_uf_updated = pd.concat( | ||
[df_ufs_latest[tb_uf_order_cols], df_uf_upstream] # type: ignore | ||
) | ||
|
||
OUTPUT_UF = os.path.join(OUTPUT, "uf.csv") | ||
|
||
df_uf_updated.to_csv(OUTPUT_UF, index=False) | ||
|
||
tb_uf.create(OUTPUT_UF, if_table_exists="replace", if_storage_data_exists="replace") |