Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into enable_cpu_on_pe
chengduoZH
7 years ago
commit
173d72b481
@ -0,0 +1,126 @@
|
||||
# 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.
|
||||
|
||||
import argparse
|
||||
|
||||
__all__ = ['parse_args', ]
|
||||
|
||||
BENCHMARK_MODELS = [
|
||||
"machine_translation", "resnet", "vgg", "mnist", "stacked_dynamic_lstm"
|
||||
]
|
||||
|
||||
|
||||
def parse_args():
|
||||
parser = argparse.ArgumentParser('Fluid model benchmarks.')
|
||||
parser.add_argument(
|
||||
'--model',
|
||||
type=str,
|
||||
choices=BENCHMARK_MODELS,
|
||||
default='resnet',
|
||||
help='The model to run benchmark with.')
|
||||
parser.add_argument(
|
||||
'--batch_size', type=int, default=32, help='The minibatch size.')
|
||||
# args related to learning rate
|
||||
parser.add_argument(
|
||||
'--learning_rate', type=float, default=0.001, help='The learning rate.')
|
||||
# TODO(wuyi): add "--use_fake_data" option back.
|
||||
parser.add_argument(
|
||||
'--skip_batch_num',
|
||||
type=int,
|
||||
default=5,
|
||||
help='The first num of minibatch num to skip, for better performance test'
|
||||
)
|
||||
parser.add_argument(
|
||||
'--iterations', type=int, default=80, help='The number of minibatches.')
|
||||
parser.add_argument(
|
||||
'--pass_num', type=int, default=100, help='The number of passes.')
|
||||
parser.add_argument(
|
||||
'--data_format',
|
||||
type=str,
|
||||
default='NCHW',
|
||||
choices=['NCHW', 'NHWC'],
|
||||
help='The data data_format, now only support NCHW.')
|
||||
parser.add_argument(
|
||||
'--device',
|
||||
type=str,
|
||||
default='GPU',
|
||||
choices=['CPU', 'GPU'],
|
||||
help='The device type.')
|
||||
parser.add_argument(
|
||||
'--gpus',
|
||||
type=int,
|
||||
default=1,
|
||||
help='If gpus > 1, will use ParallelExecutor to run, else use Executor.')
|
||||
# this option is available only for vgg and resnet.
|
||||
parser.add_argument(
|
||||
'--cpus',
|
||||
type=int,
|
||||
default=1,
|
||||
help='If cpus > 1, will use ParallelDo to run, else use Executor.')
|
||||
parser.add_argument(
|
||||
'--data_set',
|
||||
type=str,
|
||||
default='flowers',
|
||||
choices=['cifar10', 'flowers'],
|
||||
help='Optional dataset for benchmark.')
|
||||
parser.add_argument(
|
||||
'--infer_only', action='store_true', help='If set, run forward only.')
|
||||
parser.add_argument(
|
||||
'--use_cprof', action='store_true', help='If set, use cProfile.')
|
||||
parser.add_argument(
|
||||
'--use_nvprof',
|
||||
action='store_true',
|
||||
help='If set, use nvprof for CUDA.')
|
||||
parser.add_argument(
|
||||
'--no_test',
|
||||
action='store_true',
|
||||
help='If set, do not test the testset during training.')
|
||||
parser.add_argument(
|
||||
'--memory_optimize',
|
||||
action='store_true',
|
||||
help='If set, optimize runtime memory before start.')
|
||||
parser.add_argument(
|
||||
'--use_fake_data',
|
||||
action='store_true',
|
||||
help='If set ommit the actual read data operators.')
|
||||
parser.add_argument(
|
||||
'--profile', action='store_true', help='If set, profile a few steps.')
|
||||
parser.add_argument(
|
||||
'--update_method',
|
||||
type=str,
|
||||
default='local',
|
||||
choices=['local', 'pserver', 'nccl2'],
|
||||
help='Choose parameter update method, can be local, pserver, nccl2.')
|
||||
parser.add_argument(
|
||||
'--no_split_var',
|
||||
action='store_true',
|
||||
default=False,
|
||||
help='Whether split variables into blocks when update_method is pserver')
|
||||
parser.add_argument(
|
||||
'--async_mode',
|
||||
action='store_true',
|
||||
default=False,
|
||||
help='Whether start pserver in async mode to support ASGD')
|
||||
parser.add_argument(
|
||||
'--use_reader_op',
|
||||
action='store_true',
|
||||
help='Whether to use reader op, and must specify the data path if set this to true.'
|
||||
)
|
||||
parser.add_argument(
|
||||
'--data_path',
|
||||
type=str,
|
||||
default="",
|
||||
help='Directory that contains all the training recordio files.')
|
||||
args = parser.parse_args()
|
||||
return args
|
||||
File diff suppressed because it is too large
Load Diff
@ -0,0 +1,164 @@
|
||||
# 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.
|
||||
|
||||
import os
|
||||
import random
|
||||
import paddle
|
||||
import paddle.fluid as fluid
|
||||
import paddle.fluid.core as core
|
||||
from paddle.dataset import mnist, cifar, flowers, image
|
||||
|
||||
|
||||
def convert_2_recordio(py_reader, outfilepath, batch_size, shape_data,
|
||||
shape_label):
|
||||
num_batches = 0
|
||||
with fluid.program_guard(fluid.Program(), fluid.Program()):
|
||||
reader = paddle.batch(py_reader(), batch_size=batch_size)
|
||||
feeder = fluid.DataFeeder(
|
||||
feed_list=[ # order is image and label
|
||||
fluid.layers.data(
|
||||
name='image', shape=shape_data),
|
||||
fluid.layers.data(
|
||||
name='label', shape=shape_label, dtype='int64'),
|
||||
],
|
||||
place=fluid.CPUPlace())
|
||||
num_batches = fluid.recordio_writer.convert_reader_to_recordio_file(
|
||||
outfilepath, reader, feeder)
|
||||
return num_batches
|
||||
|
||||
|
||||
def prepare_mnist(outpath, batch_size):
|
||||
outfilepath = os.path.join(outpath, "mnist.recordio")
|
||||
convert_2_recordio(mnist.train, outfilepath, batch_size, [784], [1])
|
||||
|
||||
|
||||
def prepare_cifar10(outpath, batch_size):
|
||||
outfilepath = os.path.join(outpath, "cifar.recordio")
|
||||
convert_2_recordio(cifar.train10, outfilepath, batch_size, [3, 32, 32], [1])
|
||||
|
||||
|
||||
def prepare_flowers(outpath, batch_size):
|
||||
outfilepath = os.path.join(outpath, "flowers.recordio")
|
||||
convert_2_recordio(flowers.train, outfilepath, batch_size, [3, 224, 224],
|
||||
[1])
|
||||
|
||||
|
||||
def default_mapper(sample):
|
||||
img, label = sample
|
||||
img = image.simple_transform(
|
||||
img, 256, 224, True, mean=[103.94, 116.78, 123.68])
|
||||
return img.flatten().astype('float32'), label
|
||||
|
||||
|
||||
def imagenet_train(data_dir):
|
||||
contents = os.listdir(data_dir)
|
||||
if set(contents) != set(
|
||||
["train", "train.txt", "val", "val_set", "val.txt", "unzip.sh"]):
|
||||
raise Exception("Imagenet data contents error!")
|
||||
img2label = dict()
|
||||
imgfilelist = []
|
||||
with open(os.path.join(data_dir, "train.txt")) as fn:
|
||||
while 1:
|
||||
l = fn.readline()
|
||||
if not l:
|
||||
break
|
||||
img, lbl = l[:-1].split(" ")
|
||||
img2label[img] = int(lbl)
|
||||
imgfilelist.append(img)
|
||||
# shuffle all, this is slow
|
||||
random.shuffle(imgfilelist)
|
||||
|
||||
def train_reader():
|
||||
for idx, imgfile in enumerate(imgfilelist):
|
||||
data = image.load_image(
|
||||
os.path.join(data_dir, "train", imgfile.lower()))
|
||||
label = [img2label[imgfile], ]
|
||||
yield [data, label]
|
||||
|
||||
return paddle.reader.map_readers(default_mapper, train_reader)
|
||||
|
||||
|
||||
def imagenet_test(data_dir):
|
||||
contents = os.listdir(data_dir)
|
||||
if set(contents) != set(
|
||||
["train", "train.txt", "val", "val_set", "val.txt", "unzip.sh"]):
|
||||
raise Exception("Imagenet data contents error!")
|
||||
img2label = dict()
|
||||
imgfilelist = []
|
||||
with open(os.path.join(data_dir, "val.txt")) as fn:
|
||||
while 1:
|
||||
l = fn.readline()
|
||||
if not l:
|
||||
break
|
||||
img, lbl = l[:-1].split(" ")
|
||||
img2label[img] = int(lbl)
|
||||
imgfilelist.append(img)
|
||||
|
||||
def test_reader():
|
||||
for idx, imgfile in enumerate(imgfilelist):
|
||||
base_path = os.path.join(data_dir, "val", imgfile.split(".")[0])
|
||||
image_path = ".".join([base_path, "jpeg"])
|
||||
data = image.load_image(image_path)
|
||||
label = [img2label[imgfile], ]
|
||||
yield [data, label]
|
||||
|
||||
return paddle.reader.map_readers(default_mapper, test_reader)
|
||||
|
||||
|
||||
# FIXME(wuyi): delete this when https://github.com/PaddlePaddle/Paddle/pull/11066 is merged
|
||||
def convert_reader_to_recordio_files(
|
||||
filename,
|
||||
batch_per_file,
|
||||
reader_creator,
|
||||
feeder,
|
||||
compressor=core.RecordIOWriter.Compressor.Snappy,
|
||||
max_num_records=1000,
|
||||
feed_order=None):
|
||||
if feed_order is None:
|
||||
feed_order = feeder.feed_names
|
||||
f_name, f_ext = os.path.splitext(filename)
|
||||
assert (f_ext == ".recordio")
|
||||
|
||||
lines = []
|
||||
f_idx = 0
|
||||
counter = 0
|
||||
for idx, batch in enumerate(reader_creator()):
|
||||
lines.append(batch)
|
||||
if idx >= batch_per_file and idx % batch_per_file == 0:
|
||||
filename = "%s-%05d%s" % (f_name, f_idx, f_ext)
|
||||
with fluid.recordio_writer.create_recordio_writer(
|
||||
filename, compressor, max_num_records) as writer:
|
||||
for l in lines:
|
||||
res = feeder.feed(l)
|
||||
for each in feed_order:
|
||||
writer.append_tensor(res[each])
|
||||
writer.complete_append_tensor()
|
||||
counter += 1
|
||||
lines = []
|
||||
f_idx += 1
|
||||
print("written file: ", filename)
|
||||
return counter
|
||||
|
||||
|
||||
def prepare_imagenet(inpath, outpath, batch_size):
|
||||
r = paddle.batch(imagenet_train(inpath), batch_size=batch_size)
|
||||
feeder = fluid.DataFeeder(
|
||||
feed_list=[
|
||||
fluid.layers.data(
|
||||
name="image", shape=[3, 224, 224]), fluid.layers.data(
|
||||
name="label", shape=[1], dtype='int64')
|
||||
],
|
||||
place=fluid.CPUPlace())
|
||||
outpath = os.path.join(outpath, "imagenet.recordio")
|
||||
convert_reader_to_recordio_files(outpath, 10000, r, feeder)
|
||||
Some files were not shown because too many files have changed in this diff Show More
Loading…
Reference in new issue