ENH: DataFrame argument columns should accept dict/iterable type

[5j9]

Feature Type

• Adding new functionality to pandas

• Changing existing functionality in pandas

• Removing existing functionality in pandas

Problem Description

It is convenient to pass a dict to columns argument of DataFrame and use the same dict for astype conversions later. The following sample code works fine, but it fails type checking:

from pandas import DataFrame

columns = {"A": 'int64', "B": 'int32', "C": 'int16'}
df = DataFrame([[1,2,3],[4,5,6]], columns=columns)  # fails type check
df = df.astype(columns)

print(df.dtypes)

pyright output:

$ pyright 'filename.py'
filename.py
  filename.py:4:43 - error: Argument of type "dict[str, str]" cannot be assigned to parameter "columns" of type "Axes | None" in function "__init__"
    Type "dict[str, str]" cannot be assigned to type "Axes | None"
      "dict[str, str]" is incompatible with "ExtensionArray"
      "dict[str, str]" is incompatible with "ndarray[Unknown, Unknown]"
      "dict[str, str]" is incompatible with "Index"
      "dict[str, str]" is incompatible with "Series"
      "dict[str, str]" is incompatible with protocol "SequenceNotStr[Unknown]"
        "index" is not present
        "count" is not present
    ... (reportArgumentType)
1 error, 0 warnings, 0 informations

Feature Description

Change type annotations for columns to accept dict/iterable type.

Alternative Solutions

Convert columns value to a Series or numpy array or some other compatible type.

Additional Context

No response

[asishm]

do you have pandas-stubs installed?

this only reproduces without pandas-stubs for me

[5j9]

do you have pandas-stubs installed?

this only reproduces without pandas-stubs for me

I don't have pandas-stubs. You are right, works fine with stubs. I still think this should not raise error even without stubs given that pandas-stubs describes itself as "narrower than what is possibly allowed by pandas".

[twoertwein]

Would need to adjust

pandas/pandas/core/frame.py

Line 681 in 10f31f6

columns: Axes | None = None,

to include Mapping.

[phofl]

It is completely conicidential that this works and nobody should rely on it. Closing

[twoertwein]

It is completely conicidential that this works and nobody should rely on it. Closing

@Dr-Irv pandas-stubs currently declares that any dict would be okay

[Dr-Irv]

It is completely conicidential that this works and nobody should rely on it. Closing

@phofl So should we test for a dict and then raise an Exception?

[Dr-Irv]

It is completely conicidential that this works and nobody should rely on it. Closing

@Dr-Irv pandas-stubs currently declares that any dict would be okay

I checked via Blame and I put that in there at some point without a test. I just tested removing dict from Axes in pandas-stubs and it works fine, so maybe we should remove it??

[Dr-Irv]

I don't have pandas-stubs. You are right, works fine with stubs. I still think this should not raise error even without stubs given that pandas-stubs describes itself as "narrower than what is possibly allowed by pandas".

Yes, but the typing in the pandas source is much wider than what is actually allowed by pandas. Also, type checking is faster using the stubs than using the pandas source.