[C++ API Parity] [Optimizers] Merged Optimizer and LossClosureOptimizer (#34957)

Summary:
1. Removed LossClosureOptimizer, and merged Optimizer into OptimizerBase (and renamed the merged class to Optimizer)
2. Merged the LBFGS-specific serialize test function and the generic test_serialize_optimizer function.
3. BC-compatibility serialization test for LBFGS
4. Removed mentions of parameters_ in optimizer.cpp, de-virtualize all functions
5. Made defaults_ optional argument in all optimizers except SGD

**TODO**: add BC-breaking notes for this PR

Pull Request resolved: https://github.com/pytorch/pytorch/pull/34957

Differential Revision: D20645945

Pulled By: yf225

fbshipit-source-id: 383588065bf1859b38f0ad0a25d93d41e153c96e
This commit is contained in:
anjali411
2020-03-25 18:22:59 -07:00
committed by Facebook GitHub Bot
parent e08614ffd5
commit efbd6b8533
15 changed files with 150 additions and 369 deletions
+46 -10
View File
@@ -174,30 +174,66 @@ TEST(OptimTest, OptimizerAccessors) {
optimizer_.state();
}
TEST(OptimTest, BasicInterface) {
#define OLD_INTERFACE_WARNING_CHECK(func) \
{ \
std::stringstream buffer;\
torch::test::CerrRedirect cerr_redirect(buffer.rdbuf());\
func;\
ASSERT_EQ(\
torch::test::count_substr_occurrences(\
buffer.str(),\
"will be removed"\
),\
1);\
}
TEST(OptimTest, OldInterface) {
struct MyOptimizerOptions : public OptimizerCloneableOptions<MyOptimizerOptions> {
MyOptimizerOptions(double lr = 1.0) : lr_(lr) {};
TORCH_ARG(double, lr) = 1.0;
};
struct MyOptimizer : Optimizer {
using Optimizer::Optimizer;
torch::Tensor step(LossClosure closure = nullptr) override { return {};}
explicit MyOptimizer(
std::vector<at::Tensor> params, MyOptimizerOptions defaults = {}) :
Optimizer({std::move(OptimizerParamGroup(params))}, std::make_unique<MyOptimizerOptions>(defaults)) {}
};
std::vector<torch::Tensor> parameters = {
torch::ones({2, 3}), torch::zeros({2, 3}), torch::rand({2, 3})};
{
MyOptimizer optimizer(parameters);
ASSERT_EQ(optimizer.size(), parameters.size());
size_t size;
OLD_INTERFACE_WARNING_CHECK(size = optimizer.size());
ASSERT_EQ(size, parameters.size());
}
{
MyOptimizer optimizer;
ASSERT_EQ(optimizer.size(), 0);
optimizer.add_parameters(parameters);
ASSERT_EQ(optimizer.size(), parameters.size());
for (size_t p = 0; p < parameters.size(); ++p) {
ASSERT_TRUE(optimizer.parameters()[p].allclose(parameters[p]));
std::vector<at::Tensor> params;
MyOptimizer optimizer(params);
size_t size;
OLD_INTERFACE_WARNING_CHECK(size = optimizer.size());
ASSERT_EQ(size, 0);
OLD_INTERFACE_WARNING_CHECK(optimizer.add_parameters(parameters));
OLD_INTERFACE_WARNING_CHECK(size = optimizer.size());
ASSERT_EQ(size, parameters.size());
std::vector<torch::Tensor> params_;
OLD_INTERFACE_WARNING_CHECK(params_ = optimizer.parameters());
for (size_t p = 0; p < size; ++p) {
ASSERT_TRUE(params_[p].allclose(parameters[p]));
}
}
{
Linear linear(3, 4);
MyOptimizer optimizer(linear->parameters());
ASSERT_EQ(optimizer.size(), linear->parameters().size());
size_t size;
OLD_INTERFACE_WARNING_CHECK(size = optimizer.size());
ASSERT_EQ(size, linear->parameters().size());
}
}
@@ -388,7 +424,7 @@ TEST(OptimTest, AddParameter_LBFGS) {
}
LBFGS optimizer(std::vector<torch::Tensor>{}, 1.0);
optimizer.add_parameters(parameters);
OLD_INTERFACE_WARNING_CHECK(optimizer.add_parameters(parameters));
optimizer.step([]() { return torch::tensor(1); });