pandas AttributeError: can't set attribute (itertuples, shape)

AttributeError: can't set attribute in pandas almost always means you assigned to an itertuples() row (an immutable namedtuple) or to a read-only DataFrame property like df.shape or df.values. On Python 3.11+ the property case reads property 'shape' of 'DataFrame' object has no setter. Write with df.at[row.Index, col] instead.

1. The two causes: itertuples rows and DataFrame properties

import pandas as pd
df = pd.DataFrame({"name": ["a", "b"], "qty": [1, 2]})
for row in df.itertuples():
    try:
        row.qty = 10
    except AttributeError as e:
        print("itertuples row:", e)
    break
for attempt, f in [("df.shape", lambda: setattr(df, "shape", (2, 2))),
                   ("df.values", lambda: setattr(df, "values", [[0, 0]])),
                   ("df.dtypes", lambda: setattr(df, "dtypes", "int64")),
                   ("df.T", lambda: setattr(df, "T", df))]:
    try:
        f()
        print(attempt, "ok")
    except AttributeError as e:
        print(attempt + ":", e)

pandas 2.3.3 on Python 3.10.22:

/w/b1.py:9: UserWarning: Pandas doesn't allow columns to be created via a new attribute name - see https://pandas.pydata.org/pandas-docs/stable/indexing.html#attribute-access
  for attempt, f in [("df.shape", lambda: setattr(df, "shape", (2, 2))),
/w/b1.py:10: UserWarning: Pandas doesn't allow columns to be created via a new attribute name - see https://pandas.pydata.org/pandas-docs/stable/indexing.html#attribute-access
  ("df.values", lambda: setattr(df, "values", [[0, 0]])),
/w/b1.py:12: UserWarning: Pandas doesn't allow columns to be created via a new attribute name - see https://pandas.pydata.org/pandas-docs/stable/indexing.html#attribute-access
  ("df.T", lambda: setattr(df, "T", df))]:
itertuples row: can't set attribute
df.shape: can't set attribute 'shape'
df.values: can't set attribute 'values'
df.dtypes: can't set attribute 'dtypes'
df.T: can't set attribute 'T'

pandas 3.1.0 on Python 3.14.7:

/w/b1.py:9: UserWarning: Pandas doesn't allow columns to be created via a new attribute name - see https://pandas.pydata.org/docs/dev/user_guide/indexing.html#attribute-access
  for attempt, f in [("df.shape", lambda: setattr(df, "shape", (2, 2))),
/w/b1.py:10: UserWarning: Pandas doesn't allow columns to be created via a new attribute name - see https://pandas.pydata.org/docs/dev/user_guide/indexing.html#attribute-access
  ("df.values", lambda: setattr(df, "values", [[0, 0]])),
/w/b1.py:12: UserWarning: Pandas doesn't allow columns to be created via a new attribute name - see https://pandas.pydata.org/docs/dev/user_guide/indexing.html#attribute-access
  ("df.T", lambda: setattr(df, "T", df))]:
itertuples row: can't set attribute
df.shape: property 'shape' of 'DataFrame' object has no setter
df.values: property 'values' of 'DataFrame' object has no setter
df.dtypes: property 'dtypes' of 'DataFrame' object has no setter
df.T: property 'T' of 'DataFrame' object has no setter

Rows from itertuples() are namedtuples, which are immutable, so row.qty = 10 raises AttributeError: can't set attribute on both Python versions. Assigning to a read-only DataFrame property (shape, values, dtypes, T) raises the same error on Python 3.10, but Python 3.11+ rewords it as property 'shape' of 'DataFrame' object has no setter, so search results for one message may not mention the other. For three of the four, pandas also printed a UserWarning: Pandas doesn't allow columns to be created via a new attribute name just before the error.

2. Fix: write through the DataFrame with .at, or skip the loop

import pandas as pd
df = pd.DataFrame({"name": ["a", "b"], "qty": [1, 2]})
# change a cell while looping: use the index with .at
for row in df.itertuples():
    df.at[row.Index, "qty"] = row.qty * 10
print(df)
# better: no loop at all
df["qty"] = df["qty"] + 1
print(df["qty"].tolist())

Output, identical on both setups:

  name  qty
0    a   10
1    b   20
[11, 21]

row.Index is the row label, so df.at[row.Index, "qty"] = ... changes the real frame. Assigning to the row object could never have worked even if it were mutable: it is a copy. Usually the loop isn't needed at all; a vectorised column expression does the same in one line. To change dtypes use astype; to change the shape, select or reindex.

3. Related trap: df.newcol = ... does not create a column

import warnings
import pandas as pd
warnings.simplefilter("always")
df = pd.DataFrame({"qty": [1, 2]})
df.total = df["qty"] * 2
print(df.columns.tolist(), type(df.total).__name__)
df["total"] = df["qty"] * 2
print(df.columns.tolist())

pandas 2.3.3 on Python 3.10.22:

/w/b3.py:5: UserWarning: Pandas doesn't allow columns to be created via a new attribute name - see https://pandas.pydata.org/pandas-docs/stable/indexing.html#attribute-access
  df.total = df["qty"] * 2
['qty'] Series
['qty', 'total']

pandas 3.1.0 on Python 3.14.7:

/w/b3.py:5: UserWarning: Pandas doesn't allow columns to be created via a new attribute name - see https://pandas.pydata.org/docs/dev/user_guide/indexing.html#attribute-access
  df.total = df["qty"] * 2
['qty'] Series
['qty', 'total']

Assigning to an attribute name that isn't a column or property doesn't raise: it sets a plain Python attribute, warns, and the column list stays ['qty']. Always create columns with df["total"] = .... The plain-Python version of this error, for your own classes, is covered in AttributeError: can't set attribute / property has no setter.

Related: pandas 3 breaking changes · SettingWithCopyWarning · apply errors='coerce'.

Written with AI assistance. Measured on 2026-10-23 in Docker: pandas 2.3.3 on python:3.10-alpine (Python 3.10.22) and pandas 3.1.0 on python:3-alpine (Python 3.14.7). Outputs pasted unedited. PlainTables is our extension; Python and pandas are free.