Added (approximated) GELU activation function

examples/example_sine.rs

@@ -49,7 +49,7 @@ fn main() {
         Initializer::GaussianWFactor(0.0, 1.0, 0.1),
     )));
     network.add_layer(Box::new(ActivationLayer::new(
-        activation_functions::Type::LeakyRelu,
+        activation_functions::Type::Gelu,
     )));
     network.add_layer(Box::new(FCLayer::new(
         8,
@@ -57,7 +57,7 @@ fn main() {
         Initializer::GaussianWFactor(0.0, 1.0, 0.1),
     )));
     network.add_layer(Box::new(ActivationLayer::new(
-        activation_functions::Type::LeakyRelu,
+        activation_functions::Type::Gelu,
     )));
     network.add_layer(Box::new(FCLayer::new(
         1,

src/functions/activation_functions.rs

@@ -7,6 +7,7 @@ pub enum Type {
     Tanh,
     Relu,
     LeakyRelu,
+    Gelu,
 }
 
 type ActFuncTuple = (
@@ -21,6 +22,7 @@ pub fn parse_type(t: Type) -> ActFuncTuple {
         Type::Tanh => (tanh, tanh_prime),
         Type::Relu => (relu, relu_prime),
         Type::LeakyRelu => (leaky_relu, leaky_relu_prime),
+        Type::Gelu => (gelu, gelu_prime),
     }
 }
 
@@ -29,11 +31,7 @@ pub fn identity(matrix: &Array1<f64>) -> Array1<f64> {
 }
 
 pub fn identity_prime(matrix: &Array1<f64>) -> Array1<f64> {
-    let mut result = matrix.clone();
+    Array1::ones(matrix.raw_dim())
-    for x in result.iter_mut() {
-        *x = 1.0;
-    }
-    result
 }
 
 fn sigmoid(x: f64) -> f64 {
@@ -48,7 +46,7 @@ pub fn logistic(matrix: &Array1<f64>) -> Array1<f64> {
 
 pub fn logistic_prime(matrix: &Array1<f64>) -> Array1<f64> {
     let mut result = matrix.clone();
-    result.mapv_inplace(|x| sigmoid(x * (1.0 - sigmoid(x))));
+    result.mapv_inplace(|x| sigmoid(x) * (1.0 - sigmoid(x)));
     result
 }
 
@@ -87,3 +85,15 @@ pub fn leaky_relu_prime(matrix: &Array1<f64>) -> Array1<f64> {
     result.mapv_inplace(|x| if x <= 0.0 { 0.001 } else { 1.0 });
     result
 }
+
+pub fn gelu(matrix: &Array1<f64>) -> Array1<f64> {
+    let mut result = matrix.clone();
+    result.mapv_inplace(|x| x * sigmoid(1.702 * x));
+    result
+}
+
+pub fn gelu_prime(matrix: &Array1<f64>) -> Array1<f64> {
+    let mut result = matrix.clone();
+    result.mapv_inplace(|x| sigmoid(1.702 * x) * (1.0 + 1.702 * (1.0 - sigmoid(1.702 * x))));
+    result
+}