tf.keras.ops.nanmedian

Median of a tensor over the given axes, ignoring NaNs.

tf.keras.ops.nanmedian(
    x, axis=None, keepdims=False
)

This function computes the median along the specified axis or axes, skipping any NaN values. If all values along a reduced axis are NaN, the result is NaN.

x Input tensor.
axis Axis or axes along which the median is computed. If None (default), the median of the flattened tensor is returned.
keepdims If True, the reduced axes are retained as dimensions with size one. Defaults to False.

Tensor with the median values, ignoring NaNs.

Examples:

import numpy as np
from keras import ops
x = np.array([[1.0, np.nan, 3.0],
              [np.nan, 2.0, 1.0]])
ops.nanmedian(x)
1.5
 
ops.nanmedian(x, axis=1)
array([2., 1.5])
 
ops.nanmedian(x, axis=1, keepdims=True)
array([[2. ],
       [1.5]])