Fix argument documentation (#654)

* Fix argument documentation

This decay is not weight decay, so should be renamed to not be confusing.

* Change argument documentation

* Change precondition

* Update SGD.swift

Author: sgugger

Sources/TensorFlow/Optimizers/MomentumBased.swift

@@ -30,7 +30,7 @@ public class RMSProp<Model: Differentiable>: Optimizer
     public var rho: Float
     /// A small scalar added to the denominator to improve numerical stability.
     public var epsilon: Float
-    /// The weight decay.
+    /// The learning rate decay.
     public var decay: Float
     /// The step count.
     public var step: Float = 0
@@ -46,7 +46,7 @@ public class RMSProp<Model: Differentiable>: Optimizer
     ) {
         precondition(learningRate >= 0, "Learning rate must be non-negative")
         precondition(rho >= 0, "Rho must be non-negative")
-        precondition(decay >= 0, "Weight decay must be non-negative")
+        precondition(decay >= 0, "Learning rate decay must be non-negative")
 
         self.learningRate = learningRate
         self.rho = rho

Sources/TensorFlow/Optimizers/SGD.swift

@@ -25,7 +25,7 @@ public class SGD<Model: Differentiable>: Optimizer
     /// The momentum factor. It accelerates stochastic gradient descent in the relevant direction
     /// and dampens oscillations.
     public var momentum: Float
-    /// The weight decay.
+    /// The learning rate decay.
     public var decay: Float
     /// Use Nesterov momentum if true.
     public var nesterov: Bool
@@ -43,7 +43,7 @@ public class SGD<Model: Differentiable>: Optimizer
     ) {
         precondition(learningRate >= 0, "Learning rate must be non-negative")
         precondition(momentum >= 0, "Momentum must be non-negative")
-        precondition(decay >= 0, "Weight decay must be non-negative")
+        precondition(decay >= 0, "Learning rate decay must be non-negative")
 
         self.learningRate = learningRate
         self.momentum = momentum