On Wed, May 21, 2025, 4:33 PM Richard Shadrach ***@***.***> wrote: ***@***.**** requested changes on this pull request. No strong opposition to having both functions, but the performance gain of the _het version does not seem significant to me. ------------------------------ In doc/source/user_guide/indexing.rst <#61471 (comment)>: > - df = pd.DataFrame({'col': ["A", "A", "B", "B"], - 'A': [80, 23, np.nan, 22], - 'B': [80, 55, 76, 67]}) - df - idx, cols = pd.factorize(df['col']) - df.reindex(cols, axis=1).to_numpy()[np.arange(len(df)), idx] + def pd_lookup_hom(df, row_labels, col_labels): + rows = df.index.get_indexer(row_labels) Can you add df = df.loc[:, sorted(set(col_labels))] here. ------------------------------ In doc/source/user_guide/indexing.rst <#61471 (comment)>: > + +.. code-block:: python + + def pd_lookup_het(df, row_labels, col_labels): + rows = df.index.get_indexer(row_labels) + cols = df.columns.get_indexer(col_labels) + sub = df.take(np.unique(cols), axis=1) + sub = sub.take(np.unique(rows), axis=0) + rows = sub.index.get_indexer(row_labels) + values = sub.melt()["value"] + cols = sub.columns.get_indexer(col_labels) + flat_index = rows + cols * len(sub) + result = values[flat_index] + return result + +For homogeneous column types, it is fastest to skip column subsetting and go directly to numpy: Nit: NumPy ------------------------------ In doc/source/user_guide/indexing.rst <#61471 (comment)>: > -.. ipython:: python +For heterogeneous column types, we subset columns to avoid unnecessary numpy conversions: NumPy again. — Reply to this email directly, view it on GitHub <#61471 (review)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAFZOHK43SLWEMO55YCEVV327TPLNAVCNFSM6AAAAAB5TRZE7WVHI2DSMVQWIX3LMV43YUDVNRWFEZLROVSXG5CSMV3GSZLXHMZDQNJZGAZTKOJZGA> . You are receiving this because you authored the thread.Message ID: ***@***.***>