Replace native pd.to_datetime usage with convert_to_datetime

The code changes all instances of `pd.to_datetime` with a custom function `convert_to_datetime`. This function is able to correctly handle different date formats, particularly when the date information only represents a year. Additional changes include reducing explicit regex flag usage in `str.replace` and reordering some assignments for cleaner code structure.

bblocks/analysis_tools/get.py

@@ -1,7 +1,9 @@
+import datetime
 from operator import xor
 
 import pandas as pd
 
+from bblocks.cleaning_tools.clean import convert_to_datetime
 from bblocks.logger import logger
 
 
@@ -43,11 +45,11 @@ def __validate_cols(
         if col not in d.columns:
             raise ValueError(f"{col} not found in _data")
 
-    if not pd.api.types.is_datetime64_any_dtype(sdate):
-        sdate = pd.to_datetime(sdate, infer_datetime_format=True)
+    if not isinstance(sdate, datetime.datetime):
+        sdate = convert_to_datetime(sdate)
 
-    if not pd.api.types.is_datetime64_any_dtype(edate):
-        edate = pd.to_datetime(edate, infer_datetime_format=True)
+    if not isinstance(edate, datetime.datetime):
+        edate = convert_to_datetime(edate)
 
     return sdate, edate, date_col, value_col, grouper
 
@@ -81,9 +83,7 @@ def period_avg(
 
     # Check that date column is date and if not convert it
     if not pd.api.types.is_datetime64_any_dtype(data[date_column]):
-        data[date_column] = pd.to_datetime(
-            data[date_column], infer_datetime_format=True
-        )
+        data[date_column] = convert_to_datetime(data[date_column])
         logger.info(f"Converted {date_column} to datetime")
 
     # Validate args

bblocks/cleaning_tools/clean.py

@@ -3,6 +3,7 @@
 from typing import Type
 
 import country_converter as coco
+import numpy as np
 import pandas as pd
 from numpy import nan
 
@@ -90,15 +91,15 @@ def to_date_column(series: pd.Series, date_format: str | None = None) -> pd.Seri
 
     if pd.api.types.is_numeric_dtype(series):
         try:
-            return pd.to_datetime(series, format="%Y")
+            return convert_to_datetime(series)
 
         except ValueError:
             raise ValueError(
                 f"could not parse date format in. "
                 f"To fix, convert column to datetime"
             )
     if date_format is None:
-        return pd.to_datetime(series, infer_datetime_format=True)
+        return convert_to_datetime(series)
     else:
         return pd.to_datetime(series, format=date_format)
 
@@ -162,7 +163,7 @@ def date_to_str(series: pd.Series, date_format: str = "%d %B %Y") -> pd.Series:
 
     if not pd.api.types.is_datetime64_any_dtype(series):
         try:
-            series = pd.to_datetime(series, infer_datetime_format=True)
+            series = convert_to_datetime(series)
         except ValueError:
             raise ValueError(
                 f"could not parse date format in. "
@@ -240,3 +241,33 @@ def format_number(
         return series.map(formats["as_billions"].format)
 
     return series.map(f"{other_format}".format)
+
+
+def convert_to_datetime(date: str | int | pd.Series) -> pd.Series | pd.Timestamp:
+    """
+    Custom function to convert values to datetime.
+    It handles integers or strings that represent only a year.
+    """
+
+    if isinstance(date, pd.Series):
+        # Find the first non-null element in the series to determine format
+        first_valid_index = date.first_valid_index()
+        if first_valid_index is None:
+            return date.apply(lambda x: pd.NaT)
+
+        # Get the first valid value
+        format_value = date[first_valid_index]
+    else:
+        format_value = date
+
+    # Determine if the value is a year (integer or 4-digit string)
+    if isinstance(format_value, (np.integer, int)) or (
+        isinstance(format_value, str)
+        and len(format_value) == 4
+        and format_value.isdigit()
+    ):
+        format_str = "%Y"
+    else:
+        format_str = None  # Let pd.to_datetime infer the format
+
+    return pd.to_datetime(date, errors="coerce", format=format_str)

bblocks/dataframe_tools/add.py

@@ -2,7 +2,7 @@
 
 import pandas as pd
 
-from bblocks.cleaning_tools.clean import convert_id
+from bblocks.cleaning_tools.clean import convert_id, convert_to_datetime
 from bblocks.dataframe_tools.common import (
     get_population_df,
     get_poverty_ratio_df,
@@ -37,7 +37,6 @@ def __validate_add_column_params(
     df["id_"] = convert_id(df[id_column], id_type)
 
     if date_column is not None:
-
         if pd.api.types.is_numeric_dtype(df[date_column]):
             try:
                 df["merge_year"] = pd.to_datetime(df[date_column], format="%Y").dt.year
@@ -48,9 +47,7 @@ def __validate_add_column_params(
                 )
         else:
             try:
-                df["merge_year"] = pd.to_datetime(
-                    df[date_column], infer_datetime_format=True
-                ).dt.year
+                df["merge_year"] = convert_to_datetime(df[date_column]).dt.year
 
             except ValueError:
                 raise ValueError(

bblocks/dataframe_tools/common.py

@@ -2,6 +2,7 @@
 
 import pandas as pd
 
+from bblocks.cleaning_tools.clean import convert_to_datetime
 from bblocks.cleaning_tools.filter import filter_latest_by
 from bblocks.import_tools.imf import WorldEconomicOutlook
 from bblocks.import_tools.world_bank import WorldBankData
@@ -47,7 +48,6 @@ def _get_weo_indicator(
     update: bool,
     include_estimates: bool = True,
 ) -> pd.DataFrame:
-
     # Create a World Economic Outlook object
     weo = WorldEconomicOutlook().load_data(indicator=indicator)
 
@@ -57,7 +57,7 @@ def _get_weo_indicator(
 
     # Get the _data
     data = weo.get_data(keep_metadata=True).assign(
-        year=lambda d: pd.to_datetime(d.year, format="%Y")
+        year=lambda d: convert_to_datetime(d.year)
     )
 
     # Filter the _data to keep only non-estimates if needed

bblocks/import_tools/debt/common.py

@@ -7,6 +7,7 @@
 import requests
 from numpy import nan
 
+from bblocks.cleaning_tools.clean import convert_to_datetime
 from bblocks.config import BBPaths
 
 
@@ -56,7 +57,7 @@ def __clean_dsa(df: pd.DataFrame) -> pd.DataFrame:
         )
         .dropna(subset=["country"])
         .replace({"…": nan, "": nan})
-        .assign(latest_publication=lambda d: pd.to_datetime(d.latest_publication))
+        .assign(latest_publication=lambda d: convert_to_datetime(d.latest_publication))
         .reset_index(drop=True)
     )
 

bblocks/import_tools/imf.py

@@ -9,7 +9,7 @@
 from weo import all_releases, download, WEO
 
 from bblocks import config
-from bblocks.cleaning_tools.clean import clean_numeric_series
+from bblocks.cleaning_tools.clean import clean_numeric_series, convert_to_datetime
 from bblocks.import_tools.common import ImportData
 from bblocks.logger import logger
 
@@ -127,15 +127,14 @@ def __load_data(
             .rename(columns=names)
             .melt(id_vars=names.values(), var_name="year", value_name="value")
             .assign(
-                year=lambda d: pd.to_datetime(d.year, format="%Y"),
+                year=lambda d: convert_to_datetime(d.year),
                 value=lambda d: clean_numeric_series(d.value),
             )
             .dropna(subset=["value"])
             .reset_index(drop=True)
         )
 
     def _check_indicators(self, indicators: str | list | None = None) -> None | dict:
-
         if self._raw_data is None:
             self.__load_data()
 
@@ -221,7 +220,6 @@ def available_indicators(self) -> None:
     def get_data(
         self, indicators: str | list = "all", keep_metadata: bool = False
     ) -> pd.DataFrame:
-
         df = super().get_data(indicators=indicators)
 
         if not keep_metadata:

bblocks/import_tools/sdr.py

@@ -9,7 +9,7 @@
 from datetime import datetime
 import calendar
 
-from bblocks.cleaning_tools.clean import clean_numeric_series
+from bblocks.cleaning_tools.clean import clean_numeric_series, convert_to_datetime
 from bblocks.import_tools.common import ImportData
 from bblocks.config import BBPaths
 from bblocks.logger import logger
@@ -74,7 +74,7 @@ def clean_df(df: pd.DataFrame, date: str) -> pd.DataFrame:
         .rename(columns={"variable": "indicator"})
         .reset_index(drop=True)
         .assign(date=date)
-        .assign(date=lambda d: pd.to_datetime(d.date))
+        .assign(date=lambda d: convert_to_datetime(d.date))
     )
 
 

bblocks/import_tools/world_bank.py

@@ -6,6 +6,7 @@
 import pandas as pd
 import wbgapi as wb
 
+from bblocks.cleaning_tools.clean import convert_to_datetime
 from bblocks.config import BBPaths
 from bblocks.import_tools.common import ImportData
 
@@ -55,9 +56,7 @@ def _get_wb_data(
                 f"{indicator}:T": "date",
             }
         )
-        .assign(
-            indicator_code=indicator, date=lambda d: pd.to_datetime(d.date, format="%Y")
-        )
+        .assign(indicator_code=indicator, date=lambda d: convert_to_datetime(d.date))
         .sort_values(by=["iso_code", "date"])
         .reset_index(drop=True)
         .filter(["date", "iso_code", "indicator", "indicator_code", "value"], axis=1)
@@ -173,8 +172,8 @@ def clean_prices(df: pd.DataFrame) -> pd.DataFrame:
     df.columns = df.iloc[3]
     unit_dict = (
         df.iloc[4]
-        .str.replace("(", "", regex=True)
-        .str.replace(")", "", regex=True)
+        .str.replace("(", "", regex=False)
+        .str.replace(")", "", regex=False)
         .dropna()
         .to_dict()
     )
@@ -185,14 +184,16 @@ def clean_prices(df: pd.DataFrame) -> pd.DataFrame:
         .replace("..", np.nan)
         .reset_index(drop=True)
         .melt(id_vars="period", var_name="indicator", value_name="value")
-        .assign(
-            units=lambda d: d.indicator.map(unit_dict),
-            period=lambda d: pd.to_datetime(d.period, format="%YM%m"),
-            indicator=lambda d: d.indicator.str.replace(
-                "*", "", regex=True
-            ).str.strip(),
-            value=lambda d: pd.to_numeric(d.value, errors="coerce"),
-        )
+    )
+
+    df = df.assign(
+        units=lambda d: d.indicator.map(unit_dict),
+        period=lambda d: pd.to_datetime(d.period, format="%YM%m"),
+    )
+
+    df = df.assign(
+        indicator=lambda d: d.indicator.str.replace("*", "", regex=False).str.strip(),
+        value=lambda d: pd.to_numeric(d.value, errors="coerce"),
     )
 
     return df