superleaf

Readable, composable filters for pandas and plain Python data.

pip install superleaf           # numpy + pandas only
pip install "superleaf[fuzzy]"  # adds F.fuzzy_match (rapidfuzz)

Filters as data

A filter is a dict mapping columns to conditions. It can be built in one place, stored, combined and passed to dfilter:

import pandas as pd
from superleaf import F, dfilter, partition

df = pd.DataFrame({
    "type": ["T4a", "T4b", "Mi1", None, "Tm3"],
    "side": ["L", "R", "L", "R", "L"],
    "size": [10.0, None, 3.5, 7.0, -1.0],
})

left_t4 = {"type": F.startswith("T4"), "side": "L"}
dfilter(df, **left_t4)

dfilter(df, size=F.isna | F.lt(0))
dfilter(df, type=F.isin(["Mi1", "Tm3"]), columns=["type", "size"])
kept, dropped = partition(df, type=F.notna)

Keyword values are matched by equality, F.* predicates and other callables are applied to each value, and columns= selects columns by name or predicate. Use where for conditions across columns:

from superleaf import where

dfilter(df, where(lambda d: d["size"] > d["size"].median()))

Predicates work on values and columns

F predicates combine with |, & and ~, and evaluate either a single value or a whole Series with the same result:

is_valid = F.gt(0) | F.eq("n/a")
is_valid(5)                        # True
is_valid(None)                     # False: missing values only satisfy ne
is_valid.on_series(df["size"])     # boolean Series; vectorized where possible

& and | short-circuit like and and or, so a type check on the left guards the right.

The same predicate filters lists:

from superleaf.sequences import filtered, partitioned

filtered(F.startswith("T4"), ["T4a", "Mi1", "T4b"])   # ["T4a", "T4b"]

Available predicates: eq, ne, lt, le, gt, ge, isin, contains, contains_all, contains_any, startswith, endswith, startswith_one_of, endswith_one_of, matches_regex, fuzzy_match, isna, notna, isinf, notinf. Pass fallback= (and optionally exceptions=) to return a value instead of raising, e.g. F.lt(0, fallback=False)("a") is False.

Custom predicates and pipelines:

from superleaf.operators import attr_getter, bool_operator, str_op

is_even = bool_operator(lambda x: x % 2 == 0)
lower_starts_e = str_op("lower") >> F.startswith("e")   # pipe with >>
get_score = attr_getter("score") >> F.ge(0.5)

Compared with alternatives

Other utilities

• superleaf.sequences: mapped, filtered, flatten, flatmap, groupby, partitioned
• superleaf.dataframe: reorder_columns, expand_dict_to_cols
• superleaf.collections: OrderedSet, SummableDict

See the API reference and CHANGELOG.