def __init__(self, db, config):
self.feature_db = FeatureDatabase(config)
self.db = db
self.grasp_kernel = kernels.SquaredExponentialKernel(
sigma=config['kernel_sigma'], l=config['kernel_l'])
self.neighbor_kernel = kernels.SquaredExponentialKernel(
sigma=1.0, l=(1 / config['prior_neighbor_weight']))
self.neighbor_distance = config['prior_neighbor_distance']
self.num_neighbors = config['prior_num_neighbors']
self.config = config
self.grasp_kernel_tolerance = config['kernel_tolerance']
self.prior_kernel_tolerance = config['prior_kernel_tolerance']
training_datasets.append(all_training_subset_keys[:train_size])
else:
training_datasets.append(all_training_subset_keys)
if isinstance(val_size, (int)):
val_datasets.append(all_val_keys[:val_size])
else:
val_datasets.append(all_val_keys)
if isinstance(test_size, (int)):
test_datasets.append(all_test_keys[:test_size])
else:
test_datasets.append(all_test_keys)
# load all feature vectors
feature_db = FeatureDatabase(config)
all_feature_vectors = feature_db.feature_vectors()
# add training datasets
print
print 'Creating training datasets'
for training_dataset, training_size in zip(training_datasets, training_sizes):
suffix = str(training_size) + '_train'
print 'Creating set', suffix
create_index_file(training_dataset, index_file_template % (suffix))
create_nn_with_keys(all_feature_vectors, training_dataset, suffix)
# add val datasets
print
print 'Creating validation datasets'
for val_dataset, training_size in zip(val_datasets, training_sizes):
