Activations#

ivy.celu(x, /, *, alpha=1.0, complex_mode='jax', out=None)[source]#

Apply the Continuously Differentiable Exponential Linear Unit (CELU) activation function to each element of the input.

Parameters:
  • x (Union[Array, NativeArray]) – Input array.

  • alpha (float, default: 1.0) – The alpha value (negative slope) for the CELU formulation. Default is 1.0

  • complex_mode (Literal['split', 'magnitude', 'jax'], default: 'jax') – optional specifier for how to handle complex data types. See ivy.func_wrapper.handle_complex_input for more detail.

  • out (Optional[Array], default: None) – optional output array, for writing the result to. It must have a shape that the inputs broadcast to.

Return type:

Array

Returns:

ret – The input array with celu applied element-wise.

Examples

With ivy.Array input:

>>> x = ivy.array([0.39, -0.85])
>>> y = ivy.celu(x)
>>> y
ivy.array([ 0.39, -0.57])

With ivy.Container input:

>>> x = ivy.Container(a=ivy.array([0.39, -0.85]), b=ivy.array([1., -0.2]))
>>> y = ivy.celu(x)
>>> y
{
    a: ivy.array([0.38999999, -0.57]),
    b: ivy.array([1., -0.18])
}
ivy.elu(x, /, *, alpha=1.0, out=None)[source]#

Apply the elu unit function element-wise.

Parameters:
  • x (Union[Array, NativeArray]) – Input array.

  • alpha (float, default: 1.0) – scaler for controlling the slope of the function for x <= 0 Default: 1.0

  • out (Optional[Array], default: None) – optional output array, for writing the result to. It must have a shape that the inputs broadcast to.

Return type:

Array

Returns:

ret – The input array with elu applied element-wise.

Examples

With ivy.Array input: >>> x = ivy.array([0.39, -0.85]) >>> y = ivy.elu(x) >>> print(y) ivy.array([ 0.38999999, -0.57258511]) >>> x = ivy.array([1.5, 0.7, -2.4]) >>> y = ivy.zeros(3) >>> ivy.elu(x, out=y) >>> print(y) ivy.array([ 1.5, 0.69999999, -0.90928203]) >>> x = ivy.array([[1.1, 2.2, 3.3], … [-4.4, -5.5, -6.6]]) >>> ivy.elu(x, out=x) >>> print(x) ivy.array([[ 1.10000002, 2.20000005, 3.29999995],

[-0.98772264, -0.99591321, -0.99863964]])

With ivy.Container input: >>> x = ivy.Container(a=ivy.array([0.0, -1.2]), b=ivy.array([0.4, -0.2])) >>> x = ivy.elu(x, out=x) >>> print(x) {

a: ivy.array([0., -0.69880581]), b: ivy.array([0.40000001, -0.18126924])

}

ivy.hardshrink(x, /, *, lambd=0.5, out=None)[source]#

Apply the hardshrink function element-wise.

Parameters:
  • x (Union[Array, NativeArray]) – input array.

  • lambd (float, default: 0.5) – the value for the Hardshrink formulation.

  • out (Optional[Array], default: None) – optional output array, for writing the result to. It must have a shape that the inputs broadcast to.

Return type:

Array

Returns:

ret – an array containing the hardshrink activation of each element in x.

Examples

With ivy.Array input: >>> x = ivy.array([-1.0, 1.0, 2.0]) >>> y = ivy.hardshrink(x) >>> print(y) ivy.array([-1., 1., 2.]) >>> x = ivy.array([-1.0, 1.0, 2.0]) >>> y = x.hardshrink() >>> print(y) ivy.array([-1., 1., 2.]) >>> x = ivy.array([[-1.3, 3.8, 2.1], [1.7, 4.2, -6.6]]) >>> y = ivy.hardshrink(x) >>> print(y) ivy.array([[-1.29999995, 3.79999995, 2.0999999 ],

[ 1.70000005, 4.19999981, -6.5999999 ]])

ivy.hardsilu(x, /, *, out=None)[source]#

Apply the hardsilu/hardswish function element-wise.

Parameters:
  • x (Union[Array, NativeArray]) – input array

  • out (Optional[Array], default: None) – optional output array, for writing the result to. It must have a shape that the inputs broadcast to.

Return type:

Array

Returns:

an array containing the output of the hardsilu/hardswish function applied to each element in x.

Examples

With ivy.Array input:

>>> x = ivy.array([1., 2., 3.])
>>> y = ivy.hardsilu(x)
>>> print(y)
ivy.array([0.66666669, 1.66666663, 3.        ])
>>> x = ivy.array([-2.1241, 1.4897, 4.4090])
>>> y = ivy.zeros(3)
>>> ivy.hardsilu(x, out=y)
>>> print(y)
ivy.array([-0.31008321,  1.1147176 ,  4.40899992])

With ivy.Container input:

>>> x = ivy.Container(a=ivy.array([-0.5, -1, 0]), b=ivy.array([0.5, 1., 2]))
>>> y = ivy.hardsilu(x)
>>> print(y)
{
    a: ivy.array([-0.20833333, -0.33333334, 0.]),
    b: ivy.array([0.29166666, 0.66666669, 1.66666663])
}
ivy.hardtanh(x, /, *, max_val=1, min_val=-1, out=None)[source]#

Apply the hardtanh unit function element-wise.

Parameters:
  • x (Union[Array, NativeArray]) – Input array.

  • min_val (float, default: -1) – minimum value of the linear region range. Default: -1.

  • max_val (float, default: 1) – maximum value of the linear region range. Default: 1.

  • out (Optional[Array], default: None) – optional output array, for writing the result to. It must have a shape that the inputs broadcast to.

Return type:

Array

Returns:

ret – The input array with elu applied element-wise.

Examples

With ivy.Array input: >>> x = ivy.array([0.39, -0.85]) >>> y = ivy.hardtanh(x) >>> print(y) ivy.array([ 0.39, -0.85]) >>> x = ivy.array([1.5, 0.7, -2.4]) >>> y = ivy.zeros(3) >>> ivy.hardtanh(x, out=y) >>> print(y) ivy.array([ 1., 0.7, -1.]) >>> x = ivy.array([[1.1, 2.2, 3.3],[-0.4, 0.5, -6.6]]) >>> ivy.hardtanh(x, out=x) >>> print(x) ivy.array([[ 1., 1., 1.],[-0.4, 0.5, -1.]])

With ivy.Container input: >>> x = ivy.Container(a=ivy.array([0.0, -1.2]), b=ivy.array([0.4, -0.2])) >>> x = ivy.hardtanh(x, out=x) >>> print(x) {

a: ivy.array([0., -1.]), b: ivy.array([0.4, -0.2])

}

ivy.logit(x, /, *, eps=None, complex_mode='jax', out=None)[source]#

Compute the logit of x.

logit(x) = log(x / (1 - x)).

Parameters:
  • x (Union[float, int, Array]) – Input data.

  • eps (Optional[float], default: None) – When eps is None the function outputs NaN where x < 0 or x > 1. and inf or -inf where x = 1 or x = 0, respectively. Otherwise if eps is defined, x is clamped to [eps, 1 - eps]

  • complex_mode (Literal['split', 'magnitude', 'jax'], default: 'jax') – optional specifier for how to handle complex data types. See ivy.func_wrapper.handle_complex_input for more detail.

  • out (Optional[Array], default: None) – Optional output array.

Return type:

Array

Returns:

ret – Array containing elementwise logits of x.

Examples

>>> x = ivy.array([1, 0, 0.9])
>>> z = ivy.logit(x)
>>> print(z)
ivy.array([       inf,       -inf, 2.19722438])
>>> x = ivy.array([1, 2, -0.9])
>>> z = ivy.logit(x, eps=0.2)
>>> print(z)
ivy.array([ 1.38629448,  1.38629448, -1.38629436])
ivy.logsigmoid(input, /, *, complex_mode='jax', out=None)[source]#

Apply element-wise Log-sigmoid of x.

logsigmoid(x) = log(1 / (1 + exp(-x)).

Parameters:
  • input (Union[NativeArray, Array]) – Input array.

  • complex_mode (Literal['split', 'magnitude', 'jax'], default: 'jax') – optional specifier for how to handle complex data types. See ivy.func_wrapper.handle_complex_input for more detail.

Return type:

Array

Returns:

Array with same shape as input with Log-sigmoid applied to every element.

Examples

With ivy.Array input:

>>> x = ivy.array([-1., 0., 1.])
>>> z = x.logsigmoid()
>>> print(z)
ivy.array([-1.31326175, -0.69314718, -0.31326169])
>>> x = ivy.array([1.5, 0.7, -2.4])
>>> z = x.logsigmoid()
>>> print(z)
ivy.array([-0.20141329, -0.40318608, -2.48683619])

With ivy.Container input:

>>> x = ivy.Container(a=ivy.array([1.0, -1.2]), b=ivy.array([0.2, 0.6]))
>>> x = ivy.logsigmoid(x)
>>> print(x)
{
    a: ivy.array([-0.31326169, -1.46328247]),
    b: ivy.array([-0.59813893, -0.43748799])
}
ivy.prelu(x, slope, /, *, out=None)[source]#

Prelu takes input data (Array) and slope array as input,

and produces one output data (array) where the function f(x) = slope * x for x < 0, f(x) = x for x >= 0., is applied to the data array elementwise. This operator supports unidirectional broadcasting (array slope should be unidirectional broadcastable to input tensor X);

Parameters:
  • x (Union[NativeArray, Array]) – Input Array.

  • slope (Union[float, NativeArray, Array]) – Slope Array. The shape of slope can be smaller then first input X; if so, its shape must be unidirectional broadcastable to X.

  • out (Optional[Array], default: None) – Optional output array.

Return type:

Array

Returns:

ret – Array containing Parametrized relu values.

ivy.relu6(x, /, *, complex_mode='jax', out=None)[source]#

Apply the rectified linear unit 6 function element-wise.

Parameters:
  • x (Union[Array, NativeArray]) – input array

  • complex_mode (Literal['split', 'magnitude', 'jax'], default: 'jax') – optional specifier for how to handle complex data types. See ivy.func_wrapper.handle_complex_input for more detail.

  • out (Optional[Array], default: None) – optional output array, for writing the result to. It must have a shape that the inputs broadcast to.

Return type:

Array

Returns:

ret – an array containing the rectified linear unit 6 activation of each element in x.

Examples

With ivy.Array input:

>>> x = ivy.array([-1.,  0.,  1.,  2.,  3.,  4.,  5.,  6.,  7.])
>>> y = ivy.relu6(x)
>>> print(y)
ivy.array([0., 0., 1., 2., 3., 4., 5., 6., 6.])
>>> x = ivy.array([-1.,  0.,  1.,  2.,  3.,  4.,  5.,  6.,  7.])
>>> y = ivy.zeros(9)
>>> ivy.relu6(x, out = y)
>>> print(y)
ivy.array([0., 0., 1., 2., 3., 4., 5., 6., 6.])
ivy.scaled_tanh(x, /, *, alpha=1.7159, beta=0.67, out=None)[source]#

Compute the scaled hyperbolic tangent (tanh) activation.

The scaled tanh activation function is defined as: out = alpha * tanh(beta * x)

Parameters:
  • x (Union[Array, NativeArray]) – input array.

  • alpha (float, default: 1.7159) – The scaling parameter for the output. Determines the amplitude of the tanh function. Default: 1.7159

  • beta (float, default: 0.67) – The scaling parameter for the input. Determines the slope of the tanh function. Default: 0.67

  • out (Optional[Array], default: None) – optional output array, for writing the result to. It must have a shape that the inputs broadcast to.

Return type:

Array

Returns:

ret – The input array after applying the scaled tanh activation.

Examples

With ivy.Array input:

>>> x = ivy.array([22.])
>>> y = ivy.scaled_tanh(x)
>>> y
ivy.array([1.71589994]))
>>> x = ivy.array([4.0, 7.0])
>>> y = ivy.scaled_tanh(x, alpha=1.2, beta=5)
>>> y
ivy.array([1.20000005, 1.20000005])

With ivy.Container input:

>>> x = ivy.Container(a=ivy.array([1.2, -1.2]), b=ivy.array([4.4, -2.2]))
>>> y = ivy.scaled_tanh(x)
>>> y
{
    a: ivy.array([1.14324772, -1.14324772]),
    b: ivy.array([1.70648694, -1.54488957])
}
>>> x = ivy.Container(a=ivy.array([1.2]), b=ivy.array([4.4]))
>>> y = ivy.scaled_tanh(x, alpha=0.2, beta=0.5)
>>> y
{
a: ivy.array([0.10740992]),
b: ivy.array([0.19514863])
}
ivy.selu(x, /, *, out=None)[source]#

Apply the scaled exponential linear unit function element-wise.

Parameters:
  • x (Union[Array, NativeArray]) – input array

  • out (Optional[Array], default: None) – optional output array, for writing the result to. It must have a shape that the inputs broadcast to.

Return type:

Array

Returns:

ret – an array containing the scaled exponential linear unit activation of each element in x.

Examples

With ivy.Array input: >>> x = ivy.array([-1., 0., 1., 2., 3., 4., 5., 6., 7.]) >>> y = ivy.selu(x) >>> print(y) ivy.array([-1.11133075, 0. , 1.05070102, 2.10140204, 3.15210295,

4.20280409, 5.25350523, 6.30420589, 7.35490704])

>>> x = ivy.array([-1.,  0.,  1.,  2.,  3.,  4.,  5.,  6.,  7.])
>>> y = ivy.zeros(9)
>>> ivy.selu(x, out = y)
>>> print(y)
ivy.array([-1.11133075,  0.        ,  1.05070102,  2.10140204,  3.15210295,
        4.20280409,  5.25350523,  6.30420589,  7.35490704])

With ivy.Container input: >>> x = ivy.Container(a=ivy.array([-3., -2., -1., 0., 1., 2., 3., 4., 5.]), … b=ivy.array([1., 2., 3., 4., 5., 6., 7., 8., 9.]) … ) >>> x = ivy.selu(x, out=x) >>> print(x) {

a: ivy.array([-1.6705687, -1.52016652, -1.11133075, 0., 1.05070102,

2.10140204, 3.15210295, 4.20280409, 5.25350523]),

b: ivy.array([1.05070102, 2.10140204, 3.15210295, 4.20280409, 5.25350523,

6.30420589, 7.35490704, 8.40560818, 9.45630932])

}

ivy.silu(x, /, *, out=None)[source]#

Apply the silu function element-wise.

Parameters:
  • x (Union[Array, NativeArray]) – input array.

  • out (Optional[Array], default: None) – optional output array, for writing the result to. It must have a shape that the inputs broadcast to.

Return type:

Array

Returns:

ret – an array containing the silu activation of each element in x.

Examples

With ivy.Array input:

>>> x = ivy.array([-1.0, 1.0, 2.0])
>>> y = ivy.silu(x)
>>> print(y)
ivy.array([-0.2689,  0.7310,  1.7615])
>>> x = ivy.array([-1.0, 1.0, 2.0])
>>> y = x.silu()
>>> print(y)
ivy.array([-0.2689,  0.7310,  1.7615])
>>> x = ivy.array([[-1.3, 3.8, 2.1], [1.7, 4.2, -6.6]])
>>> y = ivy.silu(x)
>>> print(y)
ivy.array([[-0.2784,  3.7168,  1.8708], [ 1.4374,  4.1379, -0.0089]])
ivy.softshrink(x, /, *, lambd=0.5, out=None)[source]#

Apply the softshrink function element-wise.

Parameters:
  • x (Union[Array, NativeArray]) – input array.

  • lambd (float, default: 0.5) – the value of the lower bound of the linear region range.

  • out (Optional[Array], default: None) – optional output array, for writing the result to. It must have a shape that the inputs broadcast to.

Return type:

Array

Returns:

ret – an array containing the softshrink activation of each element in x.

Examples

With ivy.Array input: >>> x = ivy.array([-1.0, 1.0, 2.0]) >>> y = ivy.softshrink(x) >>> print(y) ivy.array([-0.5, 0.5, 1.5])

>>> x = ivy.array([-1.0, 1.0, 2.0])
>>> y = x.softshrink()
>>> print(y)
ivy.array([-0.5,  0.5,  1.5])
>>> x = ivy.array([[-1.3, 3.8, 2.1], [1.7, 4.2, -6.6]])
>>> y = ivy.softshrink(x)
>>> print(y)
ivy.array([[-0.79999995,  3.29999995,  1.59999991],
   [ 1.20000005,  3.69999981, -6.0999999 ]])
ivy.stanh(x, /, *, alpha=1.7159, beta=0.67, out=None)[source]#

Compute the scaled hyperbolic tangent (tanh) activation.

The scaled tanh activation function is defined as: out = alpha * tanh(beta * x)

Parameters:
  • x (Union[Array, NativeArray]) – input array.

  • alpha (float, default: 1.7159) – The scaling parameter for the output. Determines the amplitude of the tanh function. Default: 1.7159

  • beta (float, default: 0.67) – The scaling parameter for the input. Determines the slope of the tanh function. Default: 0.67

  • out (Optional[Array], default: None) – optional output array, for writing the result to. It must have a shape that the inputs broadcast to.

Return type:

Array

Returns:

ret – The input array after applying the scaled tanh activation.

Examples

With ivy.Array input:

>>> x = ivy.array([22.])
>>> y = ivy.scaled_tanh(x)
>>> y
ivy.array([1.71589994]))
>>> x = ivy.array([4.0, 7.0])
>>> y = ivy.scaled_tanh(x, alpha=1.2, beta=5)
>>> y
ivy.array([1.20000005, 1.20000005])

With ivy.Container input:

>>> x = ivy.Container(a=ivy.array([1.2, -1.2]), b=ivy.array([4.4, -2.2]))
>>> y = ivy.scaled_tanh(x)
>>> y
{
    a: ivy.array([1.14324772, -1.14324772]),
    b: ivy.array([1.70648694, -1.54488957])
}
>>> x = ivy.Container(a=ivy.array([1.2]), b=ivy.array([4.4]))
>>> y = ivy.scaled_tanh(x, alpha=0.2, beta=0.5)
>>> y
{
a: ivy.array([0.10740992]),
b: ivy.array([0.19514863])
}
ivy.tanhshrink(x, /, *, out=None)[source]#

Apply the tanhshrink function element-wise.

Parameters:
  • x (Union[Array, NativeArray]) – input array.

  • out (Optional[Array], default: None) – optional output array, for writing the result to. It must have a shape that the inputs broadcast to.

Return type:

Array

Returns:

ret – an array containing the tanhshrink activation of each element in x.

Examples

With ivy.Array input:

>>> x = ivy.array([-1.0, 1.0, 2.0])
>>> y = ivy.tanhshrink(x)
>>> print(y)
ivy.array([-0.23840582,  0.23840582,  1.03597236])
>>> x = ivy.array([-1.0, 1.0, 2.0])
>>> y = x.tanhshrink()
>>> print(y)
ivy.array([-0.23840582,  0.23840582,  1.03597236])
>>> x = ivy.array([[-1.3, 3.8, 2.1], [1.7, 4.2, -6.6]])
>>> y = ivy.tanhshrink(x)
>>> print(y)
ivy.array([[-0.43827677,  2.80100036,  1.12954807],
            [ 0.76459098,  3.20044947, -5.60000372]])
ivy.threshold(x, /, *, threshold, value, out=None)[source]#

Apply the threshold function element-wise.

Parameters:
  • x (Union[Array, NativeArray]) – input array.

  • threshold (float) – The value to threshold at.

  • value (float) – The value to replace with.

  • out (Optional[Array], default: None) – optional output array, for writing the result to. It must have a shape that the inputs broadcast to.

Return type:

Array

Returns:

ret – an array containing the threshold activation of each element in x.

Examples

With ivy.Array input: >>> x = ivy.array([-1.0, 1.0, 2.0]) >>> y = ivy.threshold(x,value=0.0, threshold=1.5) >>> print(y) ivy.array([0., 0., 2.])

>>> x = ivy.array([-1.0, 1.0, 2.0])
>>> x.threshold(value=0.0, threshold=1.5)
>>> print(y)
ivy.array([0., 0., 2.])
>>> x = ivy.array([[-1.3, 3.8, 2.1], [1.7, 4.2, -6.6]])
>>> y = ivy.threshold(x, value=0.0, threshold=1.5)
>>> print(y)
ivy.array([[0.        , 3.79999995, 2.0999999 ],
        [1.70000005, 4.19999981, 0.        ]])
ivy.thresholded_relu(x, /, *, threshold=0, out=None)[source]#

Apply the rectified linear unit function with custom threshold.

Parameters:
  • x (Union[Array, NativeArray]) – input array

  • threshold (Union[int, float], default: 0) – threshold value above which the activation is linear. Default: 0.

  • out (Optional[Array], default: None) – optional output array, for writing the result to. It must have a shape that the inputs broadcast to.

Return type:

Array

Returns:

ret – an array containing the rectified linear unit activation of each element in x. with custom threshold.

Examples

With ivy.Array input:

>>> x = ivy.array([-1., 0., 1.])
>>> y = ivy.thresholded_relu(x, threshold=0.5)
>>> print(y)
ivy.array([0.,  0. ,  1.])
>>> x = ivy.array([1.5, 0.7, -2.4])
>>> y = ivy.zeros(3)
>>> ivy.thresholded_relu(x, threshold=1, out = y)
>>> print(y)
ivy.array([ 1.5,  0., 0.])

With ivy.Container input:

>>> x = ivy.Container(a=ivy.array([1.0, -1.2]), b=ivy.array([0.2, 0.6]))
>>> x = ivy.thresholded_relu(x, threshold=0.5)
>>> print(x)
{
    a: ivy.array([1., 0.]),
    b: ivy.array([0., 0.6])
}