Add optimizers benchmarks

benchmarks/optimizers/__init__.py

@@ -0,0 +1,58 @@
+import six
+
+import chainer
+import cupy
+
+from benchmarks import BenchmarkBase
+
+
+class Link(chainer.Link):
+    def __init__(self, param):
+        super(Link, self).__init__()
+        with self.init_scope():
+            self.p = chainer.Parameter(param)
+
+    def __call__(self, x):
+        return x * self.p
+
+
+class OptimizerBenchmark(BenchmarkBase):
+
+    """The base class for benchmark of optimizers."""
+
+    # Call `test_*` methods only once as `backward()` has a side-effect.
+    number = 1
+
+    # Repeat the test for 10 times instead of 3 (`timeit.default_repeat`).
+    repeat = 10
+
+    def setup_benchmark(self, optimizer, batch_size, unit_num, dtype):
+        """Performs setup of benchmark for optimizers.
+
+        Call this in `setup` method of your benchmark class.
+        Note that this function performs forward computation.
+        """
+
+        xp = self.xp
+        self.optimizer = optimizer
+
+        x = xp.random.uniform(-1, 1, (batch_size, unit_num)).astype(dtype)
+        param = xp.random.uniform(-1, 1, unit_num).astype(dtype)
+        model = Link(param)
+        if xp is cupy:
+            model.to_gpu()
+
+        x = chainer.Variable(x)
+        y = model(x)
+        y.zerograd()
+        y.backward()
+        optimizer.setup(model)
+        optimizer.update()
+
+    def update(self, n_times):
+        """Runs optimizer.update()."""
+
+        optimizer = self.optimizer
+
+        for i in six.moves.range(n_times):
+            optimizer.update()

benchmarks/optimizers/ada_delta.py

@@ -0,0 +1,21 @@
+import numpy
+
+from chainer import optimizers
+
+from benchmarks.optimizers import OptimizerBenchmark
+from benchmarks.utils import backends
+from benchmarks.utils import parameterize
+
+
+@backends('gpu', 'cpu')
+@parameterize([('dtype', [numpy.float32, numpy.float64])])
+class AdaDelta(OptimizerBenchmark):
+    def setup(self, dtype):
+        unit_num = 100000
+        batch_size = 32
+        optimizer = optimizers.AdaDelta()
+
+        self.setup_benchmark(optimizer, batch_size, unit_num, dtype)
+
+    def time_update(self, dtype):
+        self.update(1000)

benchmarks/optimizers/ada_grad.py

@@ -0,0 +1,21 @@
+import numpy
+
+from chainer import optimizers
+
+from benchmarks.optimizers import OptimizerBenchmark
+from benchmarks.utils import backends
+from benchmarks.utils import parameterize
+
+
+@backends('gpu', 'cpu')
+@parameterize([('dtype', [numpy.float32, numpy.float64])])
+class AdaGrad(OptimizerBenchmark):
+    def setup(self, dtype):
+        unit_num = 100000
+        batch_size = 32
+        optimizer = optimizers.AdaGrad()
+
+        self.setup_benchmark(optimizer, batch_size, unit_num, dtype)
+
+    def time_update(self, dtype):
+        self.update(1000)

benchmarks/optimizers/adam.py

@@ -0,0 +1,23 @@
+import numpy
+
+from chainer import optimizers
+
+from benchmarks.optimizers import OptimizerBenchmark
+from benchmarks.utils import backends
+from benchmarks.utils import parameterize
+
+
+@backends('gpu', 'cpu')
+@parameterize(
+    [('dtype', [numpy.float32, numpy.float64]),
+     ('amsgrad', [True, False])])
+class Adam(OptimizerBenchmark):
+    def setup(self, dtype, amsgrad):
+        unit_num = 100000
+        batch_size = 32
+        optimizer = optimizers.Adam(0.05, amsgrad=amsgrad)
+
+        self.setup_benchmark(optimizer, batch_size, unit_num, dtype)
+
+    def time_update(self, dtype, amsgrad):
+        self.update(1000)

benchmarks/optimizers/corrected_momentum_sgd.py

@@ -0,0 +1,21 @@
+import numpy
+
+from chainer import optimizers
+
+from benchmarks.optimizers import OptimizerBenchmark
+from benchmarks.utils import backends
+from benchmarks.utils import parameterize
+
+
+@backends('gpu', 'cpu')
+@parameterize([('dtype', [numpy.float32, numpy.float64])])
+class CorrectedMomentumSGD(OptimizerBenchmark):
+    def setup(self, dtype):
+        unit_num = 100000
+        batch_size = 32
+        optimizer = optimizers.CorrectedMomentumSGD()
+
+        self.setup_benchmark(optimizer, batch_size, unit_num, dtype)
+
+    def time_update(self, dtype):
+        self.update(1000)

benchmarks/optimizers/momentum_sgd.py

@@ -0,0 +1,21 @@
+import numpy
+
+from chainer import optimizers
+
+from benchmarks.optimizers import OptimizerBenchmark
+from benchmarks.utils import backends
+from benchmarks.utils import parameterize
+
+
+@backends('gpu', 'cpu')
+@parameterize([('dtype', [numpy.float32, numpy.float64])])
+class MomentumSGD(OptimizerBenchmark):
+    def setup(self, dtype):
+        unit_num = 100000
+        batch_size = 32
+        optimizer = optimizers.MomentumSGD()
+
+        self.setup_benchmark(optimizer, batch_size, unit_num, dtype)
+
+    def time_update(self, dtype):
+        self.update(1000)

benchmarks/optimizers/msvag.py

@@ -0,0 +1,21 @@
+import numpy
+
+from chainer import optimizers
+
+from benchmarks.optimizers import OptimizerBenchmark
+from benchmarks.utils import backends
+from benchmarks.utils import parameterize
+
+
+@backends('gpu', 'cpu')
+@parameterize([('dtype', [numpy.float32, numpy.float64])])
+class MSVAG(OptimizerBenchmark):
+    def setup(self, dtype):
+        unit_num = 100000
+        batch_size = 32
+        optimizer = optimizers.MSVAG()
+
+        self.setup_benchmark(optimizer, batch_size, unit_num, dtype)
+
+    def time_update(self, dtype):
+        self.update(1000)

benchmarks/optimizers/nesterov_ag.py

@@ -0,0 +1,21 @@
+import numpy
+
+from chainer import optimizers
+
+from benchmarks.optimizers import OptimizerBenchmark
+from benchmarks.utils import backends
+from benchmarks.utils import parameterize
+
+
+@backends('gpu', 'cpu')
+@parameterize([('dtype', [numpy.float32, numpy.float64])])
+class NesterovAG(OptimizerBenchmark):
+    def setup(self, dtype):
+        unit_num = 100000
+        batch_size = 32
+        optimizer = optimizers.NesterovAG()
+
+        self.setup_benchmark(optimizer, batch_size, unit_num, dtype)
+
+    def time_update(self, dtype):
+        self.update(1000)

benchmarks/optimizers/rmsprop.py

@@ -0,0 +1,21 @@
+import numpy
+
+from chainer import optimizers
+
+from benchmarks.optimizers import OptimizerBenchmark
+from benchmarks.utils import backends
+from benchmarks.utils import parameterize
+
+
+@backends('gpu', 'cpu')
+@parameterize([('dtype', [numpy.float32, numpy.float64])])
+class RMSprop(OptimizerBenchmark):
+    def setup(self, dtype):
+        unit_num = 100000
+        batch_size = 32
+        optimizer = optimizers.RMSprop()
+
+        self.setup_benchmark(optimizer, batch_size, unit_num, dtype)
+
+    def time_update(self, dtype):
+        self.update(1000)

benchmarks/optimizers/rmsprop_graves.py

@@ -0,0 +1,21 @@
+import numpy
+
+from chainer import optimizers
+
+from benchmarks.optimizers import OptimizerBenchmark
+from benchmarks.utils import backends
+from benchmarks.utils import parameterize
+
+
+@backends('gpu', 'cpu')
+@parameterize([('dtype', [numpy.float32, numpy.float64])])
+class RMSpropGraves(OptimizerBenchmark):
+    def setup(self, dtype):
+        unit_num = 100000
+        batch_size = 32
+        optimizer = optimizers.RMSpropGraves()
+
+        self.setup_benchmark(optimizer, batch_size, unit_num, dtype)
+
+    def time_update(self, dtype):
+        self.update(1000)

benchmarks/optimizers/sgd.py

@@ -0,0 +1,21 @@
+import numpy
+
+from chainer import optimizers
+
+from benchmarks.optimizers import OptimizerBenchmark
+from benchmarks.utils import backends
+from benchmarks.utils import parameterize
+
+
+@backends('gpu', 'cpu')
+@parameterize([('dtype', [numpy.float32, numpy.float64])])
+class SGD(OptimizerBenchmark):
+    def setup(self, dtype):
+        unit_num = 100000
+        batch_size = 32
+        optimizer = optimizers.SGD()
+
+        self.setup_benchmark(optimizer, batch_size, unit_num, dtype)
+
+    def time_update(self, dtype):
+        self.update(1000)

benchmarks/optimizers/smorms3.py

@@ -0,0 +1,21 @@
+import numpy
+
+from chainer import optimizers
+
+from benchmarks.optimizers import OptimizerBenchmark
+from benchmarks.utils import backends
+from benchmarks.utils import parameterize
+
+
+@backends('gpu', 'cpu')
+@parameterize([('dtype', [numpy.float32, numpy.float64])])
+class SMORMS3(OptimizerBenchmark):
+    def setup(self, dtype):
+        unit_num = 100000
+        batch_size = 32
+        optimizer = optimizers.SMORMS3()
+
+        self.setup_benchmark(optimizer, batch_size, unit_num, dtype)
+
+    def time_update(self, dtype):
+        self.update(1000)