# init(type:a:b:) Creates an activation descriptor with the activation type and parameters a and b that you specify. ``` convenience init?(type activationType: MLCActivationType, a: Float, b: Float) ``` ## Parameters `activationType` A type of activation function. `a` Parameter a. `b` Parameter b. ## Discussion Use this initializer to create one of the following activation descriptors: - Hard Sigmoid: `f(x) = clamp((x * a) + b, 0, 1)` Activation type: [`MLCActivationType.hardSigmoid`](/documentation/MLCompute/MLCActivationType/hardSigmoid) For common behavior, set `a` to `0.2` and `b` to `0.5`. - Hyperbolic tangent (TanH): `f(x) = a * tanh(b * x)` Activation type: [`MLCActivationType.tanh`](/documentation/MLCompute/MLCActivationType/tanh) For common behavior, set `a` to `1.0` and `b` to `1.0`. - Linear: `f(x) = a * x + b` Activation type: [`MLCActivationType.linear`](/documentation/MLCompute/MLCActivationType/linear) For common behavior, set `a` to `1.0` and `b` to `0.0`. - Parametric Soft Plus: `f(x) = a * log(1 + e^(b * x))` Activation type: [`MLCActivationType.softPlus`](/documentation/MLCompute/MLCActivationType/softPlus) For common behavior, set `a` to `1.0` and `b` to `1.0`. - ReLUN: `f(x) = min((x >= 0 ? x : a * x), b)` Activation type: [`MLCActivationType.relun`](/documentation/MLCompute/MLCActivationType/relun) For common behavior, set `a` to `0.0` and `b` to `6.0`. - Threshold: `f(x) = x`, if `x > a`, else `b`, where: `a = threshold` `b = replacement` Activation type: [`MLCActivationType.threshold`](/documentation/MLCompute/MLCActivationType/threshold) --- Copyright © 2026 Apple Inc. All rights reserved. | [Terms of Use](https://www.apple.com/legal/internet-services/terms/site.html) | [Privacy Policy](https://www.apple.com/privacy/privacy-policy)