resnet_cifar.py 6.2 KB
Newer Older
Z
zhunaipan 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14
# Copyright 2019 Huawei Technologies Co., Ltd
#
# 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.
# ============================================================================
15
import argparse
Z
zhunaipan 已提交
16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50
import mindspore.nn as nn
from mindspore import Tensor
from mindspore.ops import operations as P
from mindspore.nn.optim.momentum import Momentum
from mindspore.train.model import Model, ParallelMode
from mindspore import context
import mindspore.common.dtype as mstype
import os
import numpy as np
import mindspore.ops.functional as F
from mindspore.train.callback import ModelCheckpoint, CheckpointConfig, LossMonitor
from mindspore.train.serialization import load_checkpoint, load_param_into_net
import mindspore.dataset as ds
import mindspore.dataset.transforms.c_transforms as C
import mindspore.dataset.transforms.vision.c_transforms as vision
from mindspore.communication.management import init
from mindspore.parallel._auto_parallel_context import auto_parallel_context
from resnet import resnet50
import random
random.seed(1)
np.random.seed(1)
ds.config.set_seed(1)

parser = argparse.ArgumentParser(description='Image classification')
parser.add_argument('--run_distribute', type=bool, default=False, help='Run distribute')
parser.add_argument('--device_num', type=int, default=1, help='Device num.')
parser.add_argument('--do_train', type=bool, default=True, help='Do train or not.')
parser.add_argument('--do_eval', type=bool, default=False, help='Do eval or not.')
parser.add_argument('--epoch_size', type=int, default=1, help='Epoch size.')
parser.add_argument('--batch_size', type=int, default=32, help='Batch size.')
parser.add_argument('--num_classes', type=int, default=10, help='Num classes.')
parser.add_argument('--checkpoint_path', type=str, default=None, help='Checkpoint file path')
parser.add_argument('--dataset_path', type=str, default="/var/log/npu/datasets/cifar", help='Dataset path')
args_opt = parser.parse_args()

51
device_id = int(os.getenv('DEVICE_ID'))
Z
zhunaipan 已提交
52

53
data_home = args_opt.dataset_path
Z
zhunaipan 已提交
54 55 56 57 58 59

context.set_context(mode=context.GRAPH_MODE, device_target="Ascend")
context.set_context(enable_task_sink=True, device_id=device_id)
context.set_context(enable_loop_sink=True)
context.set_context(enable_mem_reuse=True)

60

Z
zhunaipan 已提交
61 62 63 64 65 66 67
def create_dataset(repeat_num=1, training=True):
    data_dir = data_home + "/cifar-10-batches-bin"
    if not training:
        data_dir = data_home + "/cifar-10-verify-bin"
    data_set = ds.Cifar10Dataset(data_dir)

    if args_opt.run_distribute:
68 69
        rank_id = int(os.getenv('RANK_ID'))
        rank_size = int(os.getenv('RANK_SIZE'))
Z
zhunaipan 已提交
70 71 72 73 74 75 76 77
        data_set = ds.Cifar10Dataset(data_dir, num_shards=rank_size, shard_id=rank_id)

    resize_height = 224
    resize_width = 224
    rescale = 1.0 / 255.0
    shift = 0.0

    # define map operations
78
    random_crop_op = vision.RandomCrop((32, 32), (4, 4, 4, 4))  # padding_mode default CONSTANT
Z
zhunaipan 已提交
79
    random_horizontal_op = vision.RandomHorizontalFlip()
80
    resize_op = vision.Resize((resize_height, resize_width))  # interpolation default BILINEAR
Z
zhunaipan 已提交
81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106
    rescale_op = vision.Rescale(rescale, shift)
    normalize_op = vision.Normalize((0.4465, 0.4822, 0.4914), (0.2010, 0.1994, 0.2023))
    changeswap_op = vision.HWC2CHW()
    type_cast_op = C.TypeCast(mstype.int32)

    c_trans = []
    if training:
        c_trans = [random_crop_op, random_horizontal_op]
    c_trans += [resize_op, rescale_op, normalize_op,
                changeswap_op]

    # apply map operations on images
    data_set = data_set.map(input_columns="label", operations=type_cast_op)
    data_set = data_set.map(input_columns="image", operations=c_trans)

    # apply repeat operations
    data_set = data_set.repeat(repeat_num)

    # apply shuffle operations
    data_set = data_set.shuffle(buffer_size=10)

    # apply batch operations
    data_set = data_set.batch(batch_size=args_opt.batch_size, drop_remainder=True)

    return data_set

107

Z
zhunaipan 已提交
108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124
class CrossEntropyLoss(nn.Cell):
    def __init__(self):
        super(CrossEntropyLoss, self).__init__()
        self.cross_entropy = P.SoftmaxCrossEntropyWithLogits()
        self.mean = P.ReduceMean()
        self.one_hot = P.OneHot()
        self.one = Tensor(1.0, mstype.float32)
        self.zero = Tensor(0.0, mstype.float32)

    def construct(self, logits, label):
        label = self.one_hot(label, F.shape(logits)[1], self.one, self.zero)
        loss = self.cross_entropy(logits, label)[0]
        loss = self.mean(loss, (-1,))
        return loss


if __name__ == '__main__':
Z
zjun 已提交
125 126 127 128
    if not args_opt.do_eval and args_opt.run_distribute:
        context.set_auto_parallel_context(device_num=args_opt.device_num, parallel_mode=ParallelMode.DATA_PARALLEL)
        auto_parallel_context().set_all_reduce_fusion_split_indices([140])
        init()
Z
zhunaipan 已提交
129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153

    context.set_context(mode=context.GRAPH_MODE)
    epoch_size = args_opt.epoch_size
    net = resnet50(args_opt.batch_size, args_opt.num_classes)
    loss = CrossEntropyLoss()
    opt = Momentum(filter(lambda x: x.requires_grad, net.get_parameters()), 0.01, 0.9)

    model = Model(net, loss_fn=loss, optimizer=opt, metrics={'acc'})

    if args_opt.do_train:
        dataset = create_dataset(epoch_size)
        batch_num = dataset.get_dataset_size()
        config_ck = CheckpointConfig(save_checkpoint_steps=batch_num * 5, keep_checkpoint_max=10)
        ckpoint_cb = ModelCheckpoint(prefix="train_resnet_cifar10", directory="./", config=config_ck)
        loss_cb = LossMonitor()
        model.train(epoch_size, dataset, callbacks=[ckpoint_cb, loss_cb])

    if args_opt.do_eval:
        if args_opt.checkpoint_path:
            param_dict = load_checkpoint(args_opt.checkpoint_path)
            load_param_into_net(net, param_dict)
        net.set_train(False)
        eval_dataset = create_dataset(1, training=False)
        res = model.eval(eval_dataset)
        print("result: ", res)