{"id":5469,"date":"2020-11-19T17:50:57","date_gmt":"2020-11-19T12:20:57","guid":{"rendered":"http:\/\/www.pythonpool.com\/?p=5469"},"modified":"2026-07-13T12:28:52","modified_gmt":"2026-07-13T06:58:52","slug":"numpy-axis","status":"publish","type":"post","link":"https:\/\/www.pythonpool.com\/numpy-axis\/","title":{"rendered":"NumPy Axis Explained: 0, 1, Negative Axes, and keepdims"},"content":{"rendered":"

Quick answer:<\/strong> A NumPy axis identifies one dimension of an array. For a two-dimensional array, axis=0 reduces down rows and returns one value per column, while axis=1 reduces across columns and returns one value per row. Negative axes count from the end, and keepdims=True preserves a size-one dimension so the result can broadcast against the original array.<\/p>\n

\"Python
An axis identifies a dimension of an array; reductions remove that dimension unless keepdims=True preserves a broadcastable size of one.<\/figcaption><\/figure>\n

A NumPy axis is a dimension of an array. Axes are numbered from left to right in the array shape, so axis 0 is the first dimension, axis 1 is the second dimension, and so on. NumPy roll() Array Shift Guide<\/a> uses the axis parameter to shift values with wraparound along one selected dimension. Most array work should stay vectorized, but NumPy nditer() for Array Iteration<\/a> shows controlled multidimensional iteration when element-wise access is genuinely required.<\/p>\n

The official NumPy documentation defines axis<\/a> in the glossary and documents axis parameters for functions such as numpy.sum()<\/a> and numpy.expand_dims()<\/a>.<\/p>\n

The easiest way to understand axes is to look at array.shape<\/code>. For a 2D array with shape (3, 4)<\/code>, axis 0 has length 3 and axis 1 has length 4. Many NumPy operations reduce, move, or add one of those dimensions.<\/p>\n

When a function reduces an axis, that axis is the direction being collapsed. For example, sum(axis=0)<\/code> collapses rows and leaves one value for each column. sum(axis=1)<\/code> collapses columns and leaves one value for each row.<\/p>\n

That language can feel backward at first because the result is described by what remains. A good habit is to say: \u201caxis equals the dimension being operated over.\u201d Then inspect the output shape to confirm the result.<\/p>\n

Negative axes count from the end. Axis -1<\/code> means the last dimension, which is often the most convenient way to work with the final coordinate of arrays that may have different ranks.<\/p>\n

There is no universal \u201chorizontal axis\u201d or \u201cvertical axis\u201d rule that works for every array. The right interpretation depends on the array shape and the operation. For 2D arrays, axis 0 often lines up with rows and axis 1 often lines up with columns, but the reliable source is still the shape tuple. For a 2D array, NumPy fliplr: Flip Arrays Left to Right<\/a> applies the axis concept by reversing columns while preserving row order.<\/p>\n

When debugging axis code, print both the input shape and output shape. If the dimension you expected to collapse is still present, the axis value is probably wrong or keepdims=True<\/code> is preserving it intentionally.<\/p>\n

\n
\n

Contents<\/p>\nToggle<\/span><\/path><\/svg><\/svg><\/span><\/span><\/span><\/a><\/span><\/div>\n