more coolness

README.md

@@ -1,3 +1,5 @@
-# MAD
+# Marian*
 
-Massively (Parallel) Automatic Differentiation
+Parallel Automatic Differentiation Library
+
+* = in honour of Marian Rejewski, a Polish mathematician and cryptologist.

src/mad.h

@@ -40,7 +40,7 @@ class Vimpl : public Chainable {
     virtual void set_zero_adjoint() { adj_ = 0; }
     
     const Tensor& val() const { return val_; };
-    Tensor& adj() { return adj_; };
+    Tensor& grad() { return adj_; };
         
   protected:
     const Tensor val_;
@@ -72,15 +72,15 @@ class Var {
       return vimpl_->val();
     }
     
-    Tensor& adj() {
-        return vimpl_->adj();
+    Tensor& grad() {
+        return vimpl_->grad();
     }
     
     VimplPtr vimpl() const {
         return vimpl_;
     }
     
-    void grad() {
+    void calc_gradients() {
       mad::grad(vimpl_);
     }
     
@@ -99,7 +99,7 @@ struct LogVimpl : public OpVimpl {
   LogVimpl(VimplPtr a) : OpVimpl(std::log(a->val()), a) { }
   
   void chain() {
-    a_->adj() += adj_ / a_->val();
+    a_->grad() += adj_ / a_->val();
   }
 };
 
@@ -122,8 +122,8 @@ struct PlusVimplVV : public OpVimplVV {
   PlusVimplVV(VimplPtr a, VimplPtr b) : OpVimplVV(a->val() + b->val(), a, b) { }
   
   void chain() {
-    a_->adj() += adj_;
-    b_->adj() += adj_;
+    a_->grad() += adj_;
+    b_->grad() += adj_;
   }
 };
 

src/test.cpp

@@ -3,32 +3,41 @@
 
 #include "mad.h"
 
-int main(int argc, char** argv) {
+mad::Var layer(size_t max) {
     
     using namespace mad;
-    {
-        srand(time(NULL));
-        size_t max = rand() % 20 + 1;
-        
-        Var x0 = 1, x1 = 2, x2 = 3;
-        std::vector<Var> x = { x0, x1, x2 };
-        
-        Var y = 0.0;
-        for(int i = 0; i < max; i++) {
-            Var xi = i;
-            y = y + x0 + log(x2) + x1;
-            for(int j = 0; j < i; ++i) {
-                y = y + xi;
-                x.push_back(xi);
-            }
+    
+    Var x0 = 1, x1 = 2, x2 = 3;
+    Var y = 0.0;
+    for(int i = 0; i < max; i++) {
+        Var xi = i;
+        y = y + x0 + log(x2) + x1;
+        for(int j = 0; j < i; ++j) {
+            y = y + xi;
         }
-        
-           
-        set_zero_all_adjoints();
-        y.grad();
-        
-        std::cerr << "y = " << y.val() << std::endl;
-        for(int i = 0; i < x.size(); ++i)
-            std::cerr << "dy/dx_" << i << " = " << x[i].adj() << std::endl;
     }
+    
+    return y;
+}
+
+int main(int argc, char** argv) {
+    srand(time(NULL));
+    
+    using namespace mad;
+    
+    Var y1 = layer(10);
+    Var y2 = layer(5);
+    
+    Var y = y1 + log(y2);
+    
+    set_zero_all_adjoints();
+    y.calc_gradients();
+    
+    std::cerr << "y1 = " << y1.val() << std::endl;
+    std::cerr << "y2 = " << y2.val() << std::endl;
+    std::cerr << "y = " << y.val() << std::endl;
+    
+    std::cerr << "dy/dy1 = " << y1.grad() << std::endl;
+    std::cerr << "dy/dy2 = " << y2.grad() << std::endl;
+    
 }