tf.math.top_k  |  TensorFlow v2.16.1 (original) (raw)

tf.math.top_k

Stay organized with collections Save and categorize content based on your preferences.

Finds values and indices of the k largest entries for the last dimension.

View aliases

Main aliases

tf.nn.top_k

Compat aliases for migration

SeeMigration guide for more details.

tf.compat.v1.math.top_k, tf.compat.v1.nn.top_k

tf.math.top_k(
    input,
    k=1,
    sorted=True,
    index_type=tf.dtypes.int32,
    name=None
)

Used in the notebooks

If the input is a vector (rank=1), finds the k largest entries in the vector and outputs their values and indices as vectors. Thus values[j] is thej-th largest entry in input, and its index is indices[j].

result = tf.math.top_k([1, 2, 98, 1, 1, 99, 3, 1, 3, 96, 4, 1], k=3) result.values.numpy() array([99, 98, 96], dtype=int32) result.indices.numpy() array([5, 2, 9], dtype=int32)

For matrices (resp. higher rank input), computes the top k entries in each row (resp. vector along the last dimension). Thus,

input = tf.random.normal(shape=(3,4,5,6)) k = 2 values, indices = tf.math.top_k(input, k=k) values.shape.as_list() [3, 4, 5, 2] `` values.shape == indices.shape == input.shape[:-1] + [k] True

The indices can be used to gather from a tensor who's shape matches input.

gathered_values = tf.gather(input, indices, batch_dims=-1) assert tf.reduce_all(gathered_values == values)

If two elements are equal, the lower-index element appears first.

result = tf.math.top_k([1, 1, 0, 1, 0, 1, 0, 0, 1, 0, 0, 0, 1, 0], k=3) result.indices.numpy() array([0, 1, 3], dtype=int32)

By default, indices are returned as type int32, however, this can be changed by specifying the index_type.

result = tf.math.top_k([1, 1, 0, 1, 0, 1, 0, 0, 1, 0, 0, 0, 1, 0], k=3, index_type=tf.int16) result.indices.numpy() array([0, 1, 3], dtype=int16)