Learn AI Create the Future — with Pranay
Main site

Error database

ValueError: Expected 2D array, got 1D array instead (scikit-learn)

scikit-learn wants X as a table — rows by columns — even when there is one feature or one sample. Reshape with (-1, 1) for one feature, or pass a double-bracketed row.

The message you saw
ValueError: Expected 2D array, got 1D array instead (scikit-learn)

By Updated

The error

Output
ValueError: Expected 2D array, got 1D array instead:
array=[4.5 6.1 7.2].
Reshape your data either using array.reshape(-1, 1) if your data has a single feature or array.reshape(1, -1) if it contains a single sample.

A related complaint appears with scalars: Expected 2D array, got scalar array instead.

What it means

Every scikit-learn X is a 2D table: one row per sample, one column per feature. That stays true when there is only one feature, or only one sample. You passed a flat 1D array, and the library cannot tell which of the two you meant: three samples of one feature, or one sample of three features? Rather than guess, it asks you to say.

Why it happens

Selecting a single DataFrame column with single brackets gives a 1D Series:

python
X = df["experience"]        # 1D — this will fail

Predicting for one new sample with a flat list does the same:

python
model.predict([5.0])        # ambiguous 1D

How to fix it

1. One feature, many samples: reshape to a column.

python
X = df["experience"].to_numpy().reshape(-1, 1)    # shape (n, 1)
model.fit(X, y)

The -1 means "however many rows there are".

2. Cleaner with pandas: double brackets keep 2D.

python
X = df[["experience"]]      # DataFrame, shape (n, 1)

Single brackets give a Series (1D); double brackets give a DataFrame (2D). This one-character habit prevents the error entirely.

3. One sample at prediction time: wrap it in a list of lists.

python
model.predict([[5.0, 60000, 2]])     # one row, three features

The outer list is the batch, the inner list is the sample.

4. Note the direction of each reshape.

python
a.reshape(-1, 1)     # column: many samples, one feature
a.reshape(1, -1)     # row: one sample, many features

Picking the wrong one runs without error and gives nonsense results, which is worse. Check X.shape — the first number must equal your sample count.

5. Targets are the exception: y should stay 1D. If you see a DataConversionWarning about a column-vector y, flatten it with y.ravel().

How to prevent it

Print X.shape and y.shape before every fit while learning — five seconds, no surprises. Internalise the pair of rules: X is always 2D, y is usually 1D. Prefer df[["col"]] over reshape gymnastics when working from DataFrames.