pandas/removed-in-pandas-3
| Rule | removed-in-pandas-3 |
|---|---|
| Pack | pandas |
| Severity | error |
| Fix | unsafe, with xpd check --fix --unsafe-fixes |
| Library versions | every version |
These names and arguments were removed in pandas 3, so the code raises. Use the replacement that the example and the table show.
filled = df["load"].fillna(method="ffill")months = df.resample("M", on="time")["load"].sum()filled = df["load"].ffill()months = df.resample("ME", on="time")["load"].sum()What the rule reports
Section titled “What the rule reports”The rule reports applymap, fillna(method=), frequencies such as "M", "H" and "T", delim_whitespace=True, dt.week, and the removed arguments and names listed below.
How to fix it
Section titled “How to fix it”xpd check --fix --unsafe-fixes applies the fixes of this rule. Check the result, because an unsafe fix can change what the code does.
The safe fix writes map, "ME", "h", "min", a Timedelta unit such as "min" for "T", sep=r"\s+", lineterminator=, sep= for a separator passed by position, is_monotonic_increasing, np. for pd.np. and pd.json_normalize. The unsafe fix writes ffill() and bfill(), which label and order the rows of a group-by differently, "M" in resample, dt.isocalendar().week, writer.close(), pd.Series(x).value_counts(), the alias in a comparison with infer_freq or freqstr, and these forms, which raise on pandas 3: dropping delim_whitespace=False, verbose, infer_datetime_format, mangle_dupe_cols, squeeze=False, encoding of to_excel and freq of pd.Timestamp, .squeeze("columns") for squeeze=True, (s - s.mean()).abs().mean() for mad(), to_numpy() for as_matrix(), .loc for .ix, pd.Index(data, dtype="int64") for pd.Int64Index(data), from pandas import json_normalize, items(), ffill() for pad(), is_monotonic_increasing for is_lexsorted(), inclusive= for closed="left" or "right", ValueError added to a handler of the pytz errors, dt.days and dt.normalize() for a cast to a day unit. The other forms have no fix, and the comment above each of them in the rule file says why.
The rule also reports the code in the table, and each form raises on pandas 3.1.
| Kind | Reported code |
|---|---|
Arguments of read_csv |
error_bad_lines, warn_bad_lines, squeeze, verbose, keep_date_col, infer_datetime_format, mangle_dupe_cols, prefix, date_parser, and a separator passed by position. For bad lines, pass on_bad_lines="skip". |
| Writing | to_csv(line_terminator=), a separator passed to to_csv by position, to_excel(encoding=), read_excel(squeeze=), and writer.save() on a name that holds the word writer |
| Names | pd.value_counts, is_monotonic, mad, as_matrix, ix, pd.np, pd.datetime, pd.Int64Index, pd.UInt64Index, pd.Float64Index, pd.io.json.json_normalize, pandas.util.testing, SettingWithCopyWarning, the option mode.use_inf_as_na, and pd.to_numeric(errors="ignore") |
| Time series | tshift, pad(), first("3D"), last("3D"), resample(loffset=, base=, kind=), date_range(closed=), to_datetime(infer_datetime_format=), to_datetime(errors="ignore"), pct_change(fill_method="pad"), between_time(include_start=, include_end=), pd.Timestamp(freq=) and astype("timedelta64[D]") |
| Timedelta units | "T", "L", "U" and "N" in pd.Timedelta("5T"), pd.Timedelta(1, unit="T"), pd.to_timedelta(x, unit="T") and resample(offset="30T") |
| Error handlers | a handler for pytz.exceptions.AmbiguousTimeError or NonExistentTimeError, which pandas 3 no longer raises, also with a pytz zone. pandas 3 raises ValueError for a time that is repeated or missing. |
| Frequency comparisons | an old alias compared with pd.infer_freq(index) or index.freqstr, as in pd.infer_freq(index) == "H", which is never equal, because pandas returns "h" |
| Other methods | iteritems(), is_lexsorted(), lexsort_depth and groupby(axis=1) |
The rule fixes a frequency alias by what the call takes, because "M", "Q" and "Y" stay valid for periods.
| Call | What the rule does |
|---|---|
a call that takes only datetimes, such as date_range, timedelta_range, interval_range, to_offset, Grouper, floor, snap and rolling |
rewrites the alias |
resample, shift and pct_change |
rewrites "M", "Q" and "Y" with an unsafe fix, and reports "A" without a fix, because the new alias is "YE" for datetimes and "Y" for periods |
asfreq |
leaves "M" alone, and reports "A" without a fix |
to_period, period_range, pd.Period and pd.PeriodIndex, which take periods |
rewrites "H", "T", "S", "L", "U" and "N", writes "Y" for "A", and leaves "M", "Q" and "Y" alone |
The rule reports a handler for a pytz error only when its try block holds a pandas call, such as tz_localize, .dt., pd. or floor. It leaves the handler alone when the handler also names ValueError or Exception, and when the block calls the localize of pytz, which still raises the pytz errors.
The fix for an alias compared with infer_freq or freqstr is unsafe, because the comparison can then be equal. The rule reports "M" only in a comparison with infer_freq, because the freqstr of a monthly PeriodIndex is still "M". The writer.close() fix is unsafe, because xpd knows the writer only by its name. .loc reads every key as a label where ix read an integer as a position, the pytz handler then also catches a ValueError of another cause, and the fix of squeeze=True applies only to a call on one line. pandas 3 also removed DataFrame.lookup. The rule doesn't report it, because other libraries have a method of the same name.
Message
Section titled “Message”xpd prints this message with the issue. A name that starts with $ stands for code from your file.
DataFrame.applymap was removed in pandas 3. Use map, which takes the same arguments.Browse the other rules of the pandas pack.