Python chainer.optimizer() Examples

The following are 4 code examples of chainer.optimizer(). You can vote up the ones you like or vote down the ones you don't like, and go to the original project or source file by following the links above each example. You may also want to check out all available functions/classes of the module chainer , or try the search function .
Example #1
Source File: config_utils.py    From voxelnet_chainer with MIT License 5 votes vote down vote up 
def create_updater(train_iter, optimizer, config, devices):
    if "MultiprocessParallelUpdater" in config['name']:
        Updater = chainer.training.updaters.MultiprocessParallelUpdater
        return Updater(train_iter, optimizer, devices=devices,
                       converter=voxelnet_concat)

    Updater = getattr(chainer.training, config['name'])
    if "Standard" in config['name']:
        device = None if devices is None else devices['main']
        return Updater(train_iter, optimizer, device=device,
                       converter=voxelnet_concat)
    else:
        return Updater(train_iter, optimizer, devices=devices,
                       converter=voxelnet_concat)
Example #2
Source File: config_utils.py    From voxelnet_chainer with MIT License 5 votes vote down vote up 
def create_optimizer(config, model):
    Optimizer = getattr(chainer.optimizers, config['name'])
    opt = Optimizer(**config['args'])
    opt.setup(model)
    if 'hook' in config.keys():
        for key, value in config['hook'].items():
            hook = getattr(chainer.optimizer, key)
            opt.add_hook(hook(value))
    return opt
Example #3
Source File: main.py    From SPReID with MIT License 5 votes vote down vote up 
def SetupOptimizer(model):
    opt = optimizers.NesterovAG(
        lr=args.optimizer['lr'], momentum=0.9)
    opt.setup(model)
    return opt
Example #4
Source File: main.py    From SPReID with MIT License 4 votes vote down vote up 
def parse_args():
    # set extract_features to 0 for training or 1 for feature extraction
    def_extract_features = 0
    # batch size
    def_minibatch = 16
    # image size for semantic segmentation
    def_scales_tr = '512,512'
    # image size for re-identification
    def_scales_reid = '512,170'  # '778,255'
    # learning rates for fresh and pretrained layers
    def_optimizer = 'lr:0.01--lr_pretrained:0.01'
    # GPU ids
    def_GPUs = '0'
    # set checkpoint bigger than zero to load saved model from checkpoints folder
    def_checkpoint = 0
    # set pre-trained model path for finetuning using evaluation datasets
    def_model_path_for_ft = ''

    # label for the dataset
    def_dataset = 'ReID10Dx'
    # number of different ids in training data
    def_label_dim = '16803'
    def_label_dim_ft = '16803'
    # the image list for feature extraction
    def_eval_split = 'cuhk03_gallery'
    # the image list for training
    def_train_set = 'train_10d'

    # number of workers to load images parallel
    def_nb_processes = 4
    # maximum number of iterations
    def_max_iter = 200000
    # loss report interval
    def_report_interval = 50
    # number of iterations for checkpoints
    def_save_interval = 20000

    def_project_folder = '.'
    def_dataset_folder = ''
    p = ArgumentParser()
    p.add_argument('--extract_features', default=def_extract_features, type=int)
    p.add_argument('--minibatch', default=def_minibatch, type=int)
    p.add_argument('--scales_tr', default=def_scales_tr, type=str)
    p.add_argument('--scales_reid', default=def_scales_reid, type=str)
    p.add_argument('--optimizer', default=def_optimizer, type=str)
    p.add_argument('--GPUs', default=def_GPUs, type=str)
    p.add_argument('--dataset', default=def_dataset, type=str)
    p.add_argument('--eval_split', default=def_eval_split, type=str)
    p.add_argument('--train_set', default=def_train_set, type=str)
    p.add_argument('--checkpoint', default=def_checkpoint, type=int)
    p.add_argument('--model_path_for_ft', default=def_model_path_for_ft, type=str)
    p.add_argument('--label_dim', default=def_label_dim, type=str)
    p.add_argument('--label_dim_ft', default=def_label_dim_ft, type=int)
    p.add_argument('--nb_processes', default=def_nb_processes, type=int)
    p.add_argument('--max_iter', default=def_max_iter, type=int)
    p.add_argument('--report_interval', default=def_report_interval, type=int)
    p.add_argument('--save_interval', default=def_save_interval, type=int)
    p.add_argument('--project_folder', default=def_project_folder, type=str)
    p.add_argument('--dataset_folder', default=def_dataset_folder, type=str)
    args = p.parse_args()
    return args