[Relay] [Training] Allow gradient to return a tuple (apache#3600)

neo-ai · Sep 6, 2019 · 94df86f · 94df86f
1 parent 549d873
commit 94df86f
Show file tree

Hide file tree

Showing 2 changed files with 49 additions and 4 deletions.
diff --git a/src/relay/pass/gradient.cc b/src/relay/pass/gradient.cc
@@ -344,13 +344,40 @@ Expr Gradient(const Expr& re, const Module& mod) {
       args.push_back(ll->Push(Pair(p, RefCreateNode::make(ZerosLike(p)))));
     }
     auto c = ll->Push(CallNode::make(rev, args));
-    ll->Push(RefWriteNode::make(GetField(c, 1), OnesLike(GetField(c, 0))));
+    std::function<void(const Expr&, const Type&)> init_grad;
+    init_grad = [&](const Expr& e, const Type& t) {
+      if (t.as<TensorTypeNode>()) {
+        ll->Push(RefWriteNode::make(GetField(e, 1), OnesLike(GetField(e, 0))));
+      } else if (auto tt = t.as<TupleTypeNode>()) {
+        CHECK_GT(tt->fields.size(), 0);
+        init_grad(ll->Push(GetField(e, 0)), tt->fields[0]);
+      } else {
+        LOG(FATAL) << "unhandled type " << t;
+        throw;
+      }
+    };
+    init_grad(c, f->body->checked_type());
     ll->Push(CallNode::make(RefReadNode::make(bp), {}));
     std::vector<Expr> ret;
     for (const auto& a : args) {
       ret.push_back(RefReadNode::make(GetField(a, 1)));
     }
-    return Pair(GetField(c, 0), TupleNode::make(ret));
+    std::function<Expr(const Expr&, const Type&)> get_final_result;
+    get_final_result = [&](const Expr& e, const Type& t) -> Expr {
+      if (t.as<TensorTypeNode>()) {
+        return GetField(e, 0);
+      } else if (auto tt = t.as<TupleTypeNode>()) {
+        tvm::Array<Expr> fields;
+        for (size_t i = 0; i < tt->fields.size(); ++i) {
+          fields.push_back(get_final_result(ll->Push(GetField(e, i)), tt->fields[i]));
+        }
+        return TupleNode::make(fields);
+      } else {
+        LOG(FATAL) << "unhandled type " << t;
+        throw;
+      }
+    };
+    return Pair(get_final_result(c, f->body->checked_type()), TupleNode::make(ret));
   });
   return FunctionNode::make(f->params, body, GradRetType(GetRef<Function>(f)), {});
 }

diff --git a/tests/python/relay/test_pass_gradient.py b/tests/python/relay/test_pass_gradient.py
@@ -252,11 +252,28 @@ def test_if():
     net = relay.log(net)
     func = relay.Function(free_vars(net), net)
     func = run_infer_type(func)
-    net = run_infer_type(func)
-    net = gradient(net, mode='higher_order')
+    net = gradient(func, mode='higher_order')
     net = run_infer_type(net)
 
 
+def test_grad_tuple():
+    shape = (10, 10)
+    dtype = 'float32'
+    t = relay.TensorType(shape, dtype)
+    x = relay.var("x", t)
+    y = x + x
+    func = relay.Function([x], relay.Tuple([y + y, y]))
+    func = run_infer_type(func)
+    back_func = run_infer_type(gradient(func))
+    assert back_func.checked_type == relay.FuncType([t], relay.TupleType([relay.TupleType([t, t]), relay.TupleType([t])]))
+    ex = create_executor()
+    x = rand(dtype, *shape)
+    (forward_four, forward_two), (grad,) = ex.evaluate(back_func)(x)
+    tvm.testing.assert_allclose(forward_four.asnumpy(), 4 * x.asnumpy())
+    tvm.testing.assert_allclose(forward_two.asnumpy(), 2 * x.asnumpy())
+    tvm.testing.assert_allclose(grad.asnumpy(), 4 * np.ones_like(x.asnumpy()))
+
+
 if __name__ == "__main__":
     test_id()
     test_add()
@@ -269,3 +286,4 @@ def test_if():
     test_ref()
     test_square_second_order()
     test_if()
+    test_grad_tuple()