jax.numpy.isclose#

jax.numpy.isclose(a,b,rtol=1e-05,atol=1e-08,equal_nan=False)[source]#

Check if the elements of two arrays are approximately equal within a tolerance.

JAX implementation ofnumpy.allclose().

Essentially this function evaluates the following condition:

\[|a - b| \le \mathtt{atol} + \mathtt{rtol} * |b|\]

jnp.inf ina will be considered equal tojnp.inf inb.

Parameters:

• a (ArrayLike) – first input array to compare.

• b (ArrayLike) – second input array to compare.

• rtol (ArrayLike) – relative tolerance used for approximate equality. Default = 1e-05.

• atol (ArrayLike) – absolute tolerance used for approximate equality. Default = 1e-08.

• equal_nan (bool) – Boolean. IfTrue, NaNs ina will be consideredequal to NaNs inb. Default isFalse.

Returns:

A new array containing boolean values indicating whether the input arraysare element-wise approximately equal within the specified tolerances.

Return type:

Array

See also

• jax.numpy.allclose()

• jax.numpy.equal()

Examples

>>>jnp.isclose(jnp.array([1e6,2e6,jnp.inf]),jnp.array([1e6,2e7,jnp.inf]))Array([ True, False,  True], dtype=bool)>>>jnp.isclose(jnp.array([1e6,2e6,3e6]),...jnp.array([1.00008e6,2.00008e7,3.00008e8]),rtol=1e3)Array([ True,  True,  True], dtype=bool)>>>jnp.isclose(jnp.array([1e6,2e6,3e6]),...jnp.array([1.00001e6,2.00002e6,3.00009e6]),atol=1e3)Array([ True,  True,  True], dtype=bool)>>>jnp.isclose(jnp.array([jnp.nan,1,2]),...jnp.array([jnp.nan,1,2]),equal_nan=True)Array([ True,  True,  True], dtype=bool)