You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
255 lines
8.5 KiB
255 lines
8.5 KiB
# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
|
|
from __future__ import print_function
|
|
|
|
import unittest
|
|
import numpy as np
|
|
import paddle.fluid as fluid
|
|
import paddle.fluid.core as core
|
|
from paddle.fluid.op import Operator
|
|
from op_test import OpTest
|
|
import paddle
|
|
|
|
|
|
class TestSGDOp(OpTest):
|
|
def setUp(self):
|
|
self.op_type = "sgd"
|
|
self.conf()
|
|
w = np.random.random((self.h, self.w)).astype("float32")
|
|
g = np.random.random((self.h, self.w)).astype("float32")
|
|
lr = np.array([0.1]).astype("float32")
|
|
|
|
self.inputs = {'Param': w, 'Grad': g, 'LearningRate': lr}
|
|
self.outputs = {'ParamOut': w - lr * g}
|
|
|
|
def conf(self):
|
|
self.h = 102
|
|
self.w = 105
|
|
|
|
def test_check_output(self):
|
|
self.check_output()
|
|
|
|
|
|
class TestSGDOpCase8X(TestSGDOp):
|
|
def conf(self):
|
|
self.h = 10
|
|
self.w = 64
|
|
|
|
|
|
class TestSparseSGDOp(unittest.TestCase):
|
|
def check_with_place(self, place):
|
|
scope = core.Scope()
|
|
|
|
# create and initialize Grad Variable
|
|
height = 10
|
|
rows = [0, 4, 7]
|
|
self.conf()
|
|
|
|
grad_selected_rows = scope.var('Grad').get_selected_rows()
|
|
grad_selected_rows.set_height(height)
|
|
grad_selected_rows.set_rows(rows)
|
|
np_array = np.ones((len(rows), self.row_numel)).astype("float32")
|
|
np_array[0, 0] = 2.0
|
|
np_array[2, 8] = 4.0
|
|
|
|
grad_tensor = grad_selected_rows.get_tensor()
|
|
grad_tensor.set(np_array, place)
|
|
|
|
# create and initialize Param Variable
|
|
param = scope.var('Param').get_tensor()
|
|
param_array = np.full((height, self.row_numel), 5.0).astype("float32")
|
|
param.set(param_array, place)
|
|
|
|
# create and initialize LeraningRate Variable
|
|
lr = scope.var('LearningRate').get_tensor()
|
|
lr_array = np.full((1), 2.0).astype("float32")
|
|
lr.set(lr_array, place)
|
|
|
|
# create and run sgd operator
|
|
sgd_op = Operator(
|
|
"sgd",
|
|
Param='Param',
|
|
Grad='Grad',
|
|
ParamOut='Param',
|
|
LearningRate='LearningRate')
|
|
sgd_op.run(scope, place)
|
|
|
|
# get and compare result
|
|
result_array = np.array(param)
|
|
|
|
# rows[0] = 0, 5.0 - 2.0 * 2.0
|
|
self.assertAlmostEqual(1.0, result_array[rows[0], 0])
|
|
# rows[0] = 0, 5.0 - 2.0 * 1.0
|
|
self.assertAlmostEqual(3.0, result_array[rows[0], 2])
|
|
# 5.0 - 2.0 * 0.0
|
|
self.assertAlmostEqual(5.0, result_array[1, 0])
|
|
# rows[1] = 4, 5.0 - 2.0 * 1.0
|
|
self.assertAlmostEqual(3.0, result_array[rows[1], 10])
|
|
# 5.0 - 2.0 * 0.0
|
|
self.assertAlmostEqual(5.0, result_array[5, 8])
|
|
# rows[2] = 7, 5.0 - 2.0 * 1.0
|
|
self.assertAlmostEqual(3.0, result_array[rows[2], 1])
|
|
# rows[2] = 7, 5.0 - 2.0 * 4.0
|
|
self.assertAlmostEqual(-3.0, result_array[rows[2], 8])
|
|
|
|
def test_sparse_sgd(self):
|
|
places = [core.CPUPlace()]
|
|
if core.is_compiled_with_cuda():
|
|
places.append(core.CUDAPlace(0))
|
|
for place in places:
|
|
self.check_with_place(place)
|
|
|
|
def conf(self):
|
|
self.row_numel = 12
|
|
|
|
|
|
class TestSparseSGDOpCase8X(TestSparseSGDOp):
|
|
def conf(self):
|
|
self.row_numel = 16
|
|
|
|
|
|
class TestSGDOpOptimizeSelectedRows(unittest.TestCase):
|
|
def check_with_place(self, place):
|
|
scope = core.Scope()
|
|
|
|
row_width = 12
|
|
# create and initialize Grad Variable
|
|
grad_height = 10
|
|
grad_rows = [0, 4, 7]
|
|
|
|
grad_selected_rows = scope.var('Grad').get_selected_rows()
|
|
grad_selected_rows.set_height(grad_height)
|
|
grad_selected_rows.set_rows(grad_rows)
|
|
grad_array = np.ones((len(grad_rows), row_width)).astype("float32")
|
|
grad_array[0, 0] = 2.0
|
|
grad_array[2, 8] = 4.0
|
|
|
|
grad_tensor = grad_selected_rows.get_tensor()
|
|
grad_tensor.set(grad_array, place)
|
|
|
|
# create and initialize Param Variable
|
|
# create and initialize W Variable
|
|
param_rows = [0, 1, 2, 3, 4, 5, 6, 7]
|
|
|
|
# init Param
|
|
w_selected_rows = scope.var('Param').get_selected_rows()
|
|
w_selected_rows.set_height(len(param_rows))
|
|
w_selected_rows.set_rows(param_rows)
|
|
w_selected_rows.sync_index()
|
|
w_array = np.ones((len(param_rows), row_width)).astype("float32")
|
|
for i in range(len(param_rows)):
|
|
w_array[i] *= i
|
|
w_tensor = w_selected_rows.get_tensor()
|
|
w_tensor.set(w_array, place)
|
|
|
|
w_before_optimize = np.array(w_tensor)
|
|
|
|
# create and initialize LeraningRate Variable
|
|
lr_value = 0.1
|
|
lr = scope.var('LearningRate').get_tensor()
|
|
lr_array = np.full((1), lr_value).astype("float32")
|
|
lr.set(lr_array, place)
|
|
|
|
# optimize with Python
|
|
w_after_optimize = np.copy(w_before_optimize)
|
|
for index, id in enumerate(grad_rows):
|
|
w_after_optimize[id] = w_before_optimize[
|
|
id] - lr_value * grad_array[index]
|
|
|
|
# create and run sgd operator
|
|
sgd_op = Operator(
|
|
"sgd",
|
|
Param='Param',
|
|
Grad='Grad',
|
|
ParamOut='Param',
|
|
LearningRate='LearningRate')
|
|
sgd_op.run(scope, place)
|
|
|
|
# get and compare result
|
|
result_array = np.array(w_tensor)
|
|
assert (result_array == w_after_optimize).all()
|
|
|
|
def test_sparse_parameter_sgd(self):
|
|
places = [core.CPUPlace()]
|
|
# do not support GPU kernel currently
|
|
for place in places:
|
|
self.check_with_place(place)
|
|
|
|
|
|
class TestSGDOpWithLargeInput(unittest.TestCase):
|
|
def runTest(self):
|
|
data = fluid.layers.fill_constant(shape=[1], value=128, dtype='int64')
|
|
label = fluid.layers.fill_constant(
|
|
shape=[1, 150], value=0.5, dtype='float32')
|
|
emb = fluid.embedding(input=data, size=(10000000, 150), dtype='float32')
|
|
out = fluid.layers.l2_normalize(x=emb, axis=-1)
|
|
|
|
cost = fluid.layers.square_error_cost(input=out, label=label)
|
|
avg_cost = fluid.layers.mean(cost)
|
|
sgd_optimizer = fluid.optimizer.SGD(learning_rate=0.001)
|
|
sgd_optimizer.minimize(avg_cost)
|
|
|
|
place = fluid.CPUPlace()
|
|
exe = fluid.Executor(place)
|
|
exe.run(fluid.default_startup_program())
|
|
compiled_prog = fluid.compiler.CompiledProgram(
|
|
fluid.default_main_program())
|
|
result = exe.run(compiled_prog, fetch_list=[avg_cost])
|
|
|
|
|
|
class TestSGDV2(unittest.TestCase):
|
|
def test_sgd_dygraph(self):
|
|
paddle.disable_static()
|
|
value = np.arange(26).reshape(2, 13).astype("float32")
|
|
a = paddle.to_tensor(value)
|
|
linear = paddle.nn.Linear(13, 5)
|
|
# This can be any optimizer supported by dygraph.
|
|
adam = paddle.optimizer.SGD(learning_rate=0.01,
|
|
parameters=linear.parameters(),
|
|
weight_decay=0.01)
|
|
out = linear(a)
|
|
out.backward()
|
|
adam.step()
|
|
adam.clear_gradients()
|
|
|
|
def test_sgd(self):
|
|
place = fluid.CPUPlace()
|
|
main = fluid.Program()
|
|
with fluid.program_guard(main):
|
|
x = fluid.layers.data(name='x', shape=[13], dtype='float32')
|
|
y = fluid.layers.data(name='y', shape=[1], dtype='float32')
|
|
y_predict = fluid.layers.fc(input=x, size=1, act=None)
|
|
cost = fluid.layers.square_error_cost(input=y_predict, label=y)
|
|
avg_cost = fluid.layers.mean(cost)
|
|
|
|
rms_optimizer = paddle.optimizer.SGD(learning_rate=0.1)
|
|
rms_optimizer.minimize(avg_cost)
|
|
|
|
fetch_list = [avg_cost]
|
|
train_reader = paddle.batch(
|
|
paddle.dataset.uci_housing.train(), batch_size=1)
|
|
feeder = fluid.DataFeeder(place=place, feed_list=[x, y])
|
|
exe = fluid.Executor(place)
|
|
exe.run(fluid.default_startup_program())
|
|
for data in train_reader():
|
|
exe.run(main, feed=feeder.feed(data), fetch_list=fetch_list)
|
|
|
|
def test_raise_error(self):
|
|
self.assertRaises(ValueError, paddle.optimizer.SGD, learning_rate=None)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|