ValueError: setting an array element with a sequence. — how to fix

Traceback (most recent call last):
  File "app.py", line 4, in <module>
    arr = np.array([[1, 2, 3], [4, 5]], dtype=float)
ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detected shape was (2,) + inhomogeneous part.

Quick fix: Check the lengths: print({len(r) for r in rows}). Pad or truncate rows to equal length (or use dtype=object if you really want ragged data). When assigning, match shapes: arr[i, :] = row instead of arr[i] = [row, extra]. For pandas columns of lists, use np.stack(df['col'].to_numpy()).

What it means

A NumPy array with a numeric dtype holds exactly one number per element, and all rows must have the same length. If NumPy finds a nested list where it expects a scalar, it raises this error. NumPy 1.24+ adds The requested array has an inhomogeneous shape after N dimensions, which tells you the depth at which lengths stopped matching.

Cause 1: ragged nested lists

rows = [[1, 2, 3], [4, 5]]
np.array(rows)                 # ValueError (NumPy >= 1.24)

Find the bad rows:

from collections import Counter
print(Counter(len(r) for r in rows))
print([i for i, r in enumerate(rows) if len(r) != 3])

Fix by padding to a fixed length:

width = max(len(r) for r in rows)
arr = np.array([r + [0] * (width - len(r)) for r in rows], dtype=float)

Or, for variable-length sequences that you won't do math on, keep them as objects:

arr = np.empty(len(rows), dtype=object)
arr[:] = rows

In older NumPy (< 1.24) ragged input silently became dtype=object with a VisibleDeprecationWarning; code that "worked" before may fail after an upgrade for this reason.

Cause 2: assigning a sequence to a single element

a = np.zeros(3)
a[0] = [1, 2]                  # one slot, two values

Assign to a slice of matching shape, or make the array 2-D:

a = np.zeros((3, 2))
a[0] = [1, 2]                  # row 0 now [1., 2.]

Cause 3: pandas column containing lists or arrays

Feeding df["embedding"] (each cell a list) into scikit-learn, np.array(df.values, dtype=float) or df.astype(float) triggers it. Stack the cells into a 2-D array:

X = np.stack(df["embedding"].to_numpy())          # all cells same length
# or expand into columns
emb = pd.DataFrame(df["embedding"].tolist(), index=df.index)

If the cells are strings like "[0.1, 0.2]" (after a CSV round-trip), parse them first: df["embedding"].apply(json.loads).

Cause 4: mixing scalars and arrays

np.array([1.0, np.array([2.0, 3.0])], dtype=float)  # ValueError

Flatten first: np.concatenate([[1.0], np.array([2.0, 3.0])]) or np.hstack(...).

Cause 5: functions returning different shapes

When building arrays in a loop, one iteration may return a shorter result (e.g. an empty list on error). Assert shape as you go:

features = []
for item in items:
    f = extract(item)
    assert len(f) == 128, (item, len(f))
    features.append(f)
X = np.array(features)

Better still, preallocate: X = np.zeros((len(items), 128)) and fill X[i] = extract(item); the bad row raises immediately with a clear broadcast error.

Last updated 2026-11-05.