def main(_): # Please make sure that a root dir is specified before running this script! root_dir = None model_config = mc.ModelConfig(model_type='nin', dataset='cifar10') load_and_run(model_config, root_dir) print('Loaded a NIN_CIFAR10 model.') # example for resnet cifar100 model_config = mc.ModelConfig(model_type='resnet', dataset='cifar100') load_and_run(model_config, root_dir) print('Loaded a RESNET_CIFAR100 model.')
def main(_): # Please make sure that a root dir is specified before running this script! root_dir = None model_config = mc.ModelConfig( model_type='nin', dataset='cifar10', root_dir=root_dir) load_and_run(model_config, root_dir) print('Loaded a NIN_CIFAR10 model.') print('Evaluating the NIN_CIFAR10 model.') eval_result = evaluate_model(model_config, root_dir) print('Test Accuracy: {}'.format(eval_result)) print('Stored Test Accuracy: {}'.format(model_config.test_stats())) print('Stored Train Accuracy: {}'.format(model_config.training_stats())) print('==========================================') # example for resnet cifar100 model_config = mc.ModelConfig(model_type='resnet', dataset='cifar100') load_and_run(model_config, root_dir) print('Loaded a RESNET_CIFAR100 model.')
count = 0 for wide_multiplier in d[key]['wide_multiplier']: for batchnorm in d[key]['batchnorm']: for dropout_prob in d[key]['dropout_prob']: for augmentation in d[key]['augmentation']: for decay_fac in d[key]['decay_fac']: for copy in d[key]['copy']: for normalization in d[key]['normalization']: for learning_rate in d[key]['learning_rate']: model_config = mc.ModelConfig( model_type=model_type, dataset=dataset, wide_multiplier=wide_multiplier, batchnorm=batchnorm, dropout_prob=dropout_prob, data_augmentation=augmentation, l2_decay_factor=decay_fac, normalization=normalization, learning_rate=learning_rate, copy=copy, root_dir=root_dir) count += 1 train_loss = np.NaN train_cross_entropy = np.NaN train_global_step = np.NaN train_accuracy = np.NaN eval_loss = np.NaN eval_cross_entropy = np.NaN eval_global_step = np.NaN eval_accuracy = np.NaN try: