Python utils.Foo方法代碼示例

# 需要導入模塊: from src import utils [as 別名]
# 或者: from src.utils import Foo [as 別名]
def get_arch_vars(arch_str):
  if arch_str == '': vals = []
  else: vals = arch_str.split('_')
  
  ks = ['ver', 'lstm_dim', 'dropout']
  
  # Exp Ver
  if len(vals) == 0: vals.append('v0')
  # LSTM dimentsions
  if len(vals) == 1: vals.append('lstm2048')
  # Dropout
  if len(vals) == 2: vals.append('noDO')
  
  assert(len(vals) == 3)
  
  vars = utils.Foo()
  for k, v in zip(ks, vals):
    setattr(vars, k, v)
  
  logging.error('arch_vars: %s', vars)
  return vars 

# 需要導入模塊: from src import utils [as 別名]
# 或者: from src.utils import Foo [as 別名]
def get_arch_vars(arch_str):
  if arch_str == '': vals = []
  else: vals = arch_str.split('_')
  ks = ['var1', 'var2', 'var3']
  ks = ks[:len(vals)]
  
  # Exp Ver.
  if len(vals) == 0: ks.append('var1'); vals.append('v0')
  # custom arch.
  if len(vals) == 1: ks.append('var2'); vals.append('')
  # map scape for projection baseline.
  if len(vals) == 2: ks.append('var3'); vals.append('fr2')

  assert(len(vals) == 3)

  vars = utils.Foo()
  for k, v in zip(ks, vals):
    setattr(vars, k, v)

  logging.error('arch_vars: %s', vars)
  return vars 

# 需要導入模塊: from src import utils [as 別名]
# 或者: from src.utils import Foo [as 別名]
def _write_map_files(b_in, b_out, transform):
  cats = get_categories()

  env = utils.Foo(padding=10, resolution=5, num_point_threshold=2,
                  valid_min=-10, valid_max=200, n_samples_per_face=200)
  robot = utils.Foo(radius=15, base=10, height=140, sensor_height=120,
                    camera_elevation_degree=-15)
  
  building_loader = factory.get_dataset('sbpd')
  for flip in [False, True]:
    b = nav_env.Building(b_out, robot, env, flip=flip,
                         building_loader=building_loader)
    logging.info("building_in: %s, building_out: %s, transform: %d", b_in,
                 b_out, transform)
    maps = _get_semantic_maps(b_in, transform, b.map, flip, cats)
    maps = np.transpose(np.array(maps), axes=[1,2,0])

    #  Load file from the cache.
    file_name = '{:s}_{:d}_{:d}_{:d}_{:d}_{:d}_{:d}.pkl'
    file_name = file_name.format(b.building_name, b.map.size[0], b.map.size[1],
                                 b.map.origin[0], b.map.origin[1],
                                 b.map.resolution, flip)
    out_file = os.path.join(DATA_DIR, 'processing', 'class-maps', file_name)
    logging.info('Writing semantic maps to %s.', out_file)
    save_variables(out_file, [maps, cats], ['maps', 'cats'], overwrite=True) 

# 需要導入模塊: from src import utils [as 別名]
# 或者: from src.utils import Foo [as 別名]
def get_default_summary_ops():
  return utils.Foo(summary_ops=None, print_summary_ops=None, 
                   additional_return_ops=[], arop_summary_iters=[],
                   arop_eval_fns=[]) 

# 需要導入模塊: from src import utils [as 別名]
# 或者: from src.utils import Foo [as 別名]
def adjust_args_for_mode(args, mode):
  if mode == 'train':
    args.control.train = True
  
  elif mode == 'val1':
    # Same settings as for training, to make sure nothing wonky is happening
    # there.
    args.control.test = True
    args.control.test_mode = 'val'
    args.navtask.task_params.batch_size = 32

  elif mode == 'val2':
    # No data augmentation, not sampling but taking the argmax action, not
    # sampling from the ground truth at all.
    args.control.test = True
    args.arch.action_sample_type = 'argmax'
    args.arch.sample_gt_prob_type = 'zero'
    args.navtask.task_params.data_augment = \
      utils.Foo(lr_flip=0, delta_angle=0, delta_xy=0, relight=False,
                relight_fast=False, structured=False)
    args.control.test_mode = 'val'
    args.navtask.task_params.batch_size = 32

  elif mode == 'bench':
    # Actually testing the agent in settings that are kept same between
    # different runs.
    args.navtask.task_params.batch_size = 16
    args.control.test = True
    args.arch.action_sample_type = 'argmax'
    args.arch.sample_gt_prob_type = 'zero'
    args.navtask.task_params.data_augment = \
      utils.Foo(lr_flip=0, delta_angle=0, delta_xy=0, relight=False,
                relight_fast=False, structured=False)
    args.summary.test_iters = 250
    args.control.only_eval_when_done = True
    args.control.reset_rng_seed = True
    args.control.test_mode = 'test'
  else:
    logging.fatal('Unknown mode: %s.', mode)
    assert(False)
  return args 

# 需要導入模塊: from src import utils [as 別名]
# 或者: from src.utils import Foo [as 別名]
def get_solver_vars(solver_str):
  if solver_str == '': vals = []; 
  else: vals = solver_str.split('_')
  ks = ['clip', 'dlw', 'long', 'typ', 'isdk', 'adam_eps', 'init_lr'];
  ks = ks[:len(vals)]

  # Gradient clipping or not.
  if len(vals) == 0: ks.append('clip'); vals.append('noclip');
  # data loss weight.
  if len(vals) == 1: ks.append('dlw');  vals.append('dlw20')
  # how long to train for.
  if len(vals) == 2: ks.append('long');  vals.append('nolong')
  # Adam
  if len(vals) == 3: ks.append('typ');  vals.append('adam2')
  # reg loss wt
  if len(vals) == 4: ks.append('rlw');  vals.append('rlw1')
  # isd_k
  if len(vals) == 5: ks.append('isdk');  vals.append('isdk415') # 415, inflexion at 2.5k.
  # adam eps
  if len(vals) == 6: ks.append('adam_eps');  vals.append('aeps1en8')
  # init lr
  if len(vals) == 7: ks.append('init_lr');  vals.append('lr1en3')

  assert(len(vals) == 8)
  
  vars = utils.Foo()
  for k, v in zip(ks, vals):
    setattr(vars, k, v)
  logging.error('solver_vars: %s', vars)
  return vars 

# 需要導入模塊: from src import utils [as 別名]
# 或者: from src.utils import Foo [as 別名]
def get_navtask_vars(navtask_str):
  if navtask_str == '': vals = []
  else: vals = navtask_str.split('_')

  ks_all = ['dataset_name', 'modality', 'task', 'history', 'max_dist',
            'num_steps', 'step_size', 'n_ori', 'aux_views', 'data_aug']
  ks = ks_all[:len(vals)]

  # All data or not.
  if len(vals) == 0: ks.append('dataset_name'); vals.append('sbpd')
  # modality
  if len(vals) == 1: ks.append('modality'); vals.append('rgb')
  # semantic task?
  if len(vals) == 2: ks.append('task'); vals.append('r2r')
  # number of history frames.
  if len(vals) == 3: ks.append('history'); vals.append('h0')
  # max steps
  if len(vals) == 4: ks.append('max_dist'); vals.append('32')
  # num steps
  if len(vals) == 5: ks.append('num_steps'); vals.append('40')
  # step size
  if len(vals) == 6: ks.append('step_size'); vals.append('8')
  # n_ori
  if len(vals) == 7: ks.append('n_ori'); vals.append('4')
  # Auxiliary views.
  if len(vals) == 8: ks.append('aux_views'); vals.append('nv0')
  # Normal data augmentation as opposed to structured data augmentation (if set
  # to straug.
  if len(vals) == 9: ks.append('data_aug'); vals.append('straug')

  assert(len(vals) == 10)
  for i in range(len(ks)):
    assert(ks[i] == ks_all[i])

  vars = utils.Foo()
  for k, v in zip(ks, vals):
    setattr(vars, k, v)
  logging.error('navtask_vars: %s', vals)
  return vars 

# 需要導入模塊: from src import utils [as 別名]
# 或者: from src.utils import Foo [as 別名]
def get_default_args():
  summary_args = utils.Foo(display_interval=1, test_iters=26,
                           arop_full_summary_iters=14)

  control_args = utils.Foo(train=False, test=False,
                           force_batchnorm_is_training_at_test=False,
                           reset_rng_seed=False, only_eval_when_done=False,
                           test_mode=None)
  return summary_args, control_args 

# 需要導入模塊: from src import utils [as 別名]
# 或者: from src.utils import Foo [as 別名]
def get_args_for_config(config_name):
  args = utils.Foo()

  args.summary, args.control = get_default_args()

  exp_name, mode_str = config_name.split('+')
  arch_str, solver_str, navtask_str = exp_name.split('.')
  logging.error('config_name: %s', config_name)
  logging.error('arch_str: %s', arch_str)
  logging.error('navtask_str: %s', navtask_str)
  logging.error('solver_str: %s', solver_str)
  logging.error('mode_str: %s', mode_str)

  args.solver = cc.process_solver_str(solver_str)
  args.navtask = cc.process_navtask_str(navtask_str)

  args = process_arch_str(args, arch_str)
  args.arch.isd_k = args.solver.isd_k

  # Train, test, etc.
  mode, imset = mode_str.split('_')
  args = cc.adjust_args_for_mode(args, mode)
  args.navtask.building_names = args.navtask.dataset.get_split(imset)
  args.control.test_name = '{:s}_on_{:s}'.format(mode, imset)

  # Log the arguments
  logging.error('%s', args)
  return args 

# 需要導入模塊: from src import utils [as 別名]
# 或者: from src.utils import Foo [as 別名]
def get_default_cmp_args():
  batch_norm_param = {'center': True, 'scale': True,
                      'activation_fn':tf.nn.relu}

  mapper_arch_args = utils.Foo(
      dim_reduce_neurons=64,
      fc_neurons=[1024, 1024],
      fc_out_size=8,
      fc_out_neurons=64,
      encoder='resnet_v2_50',
      deconv_neurons=[64, 32, 16, 8, 4, 2],
      deconv_strides=[2, 2, 2, 2, 2, 2],
      deconv_layers_per_block=2,
      deconv_kernel_size=4,
      fc_dropout=0.5,
      combine_type='wt_avg_logits',
      batch_norm_param=batch_norm_param)

  readout_maps_arch_args = utils.Foo(
      num_neurons=[],
      strides=[],
      kernel_size=None,
      layers_per_block=None)

  arch_args = utils.Foo(
      vin_val_neurons=8, vin_action_neurons=8, vin_ks=3, vin_share_wts=False,
      pred_neurons=[64, 64], pred_batch_norm_param=batch_norm_param,
      conv_on_value_map=0, fr_neurons=16, fr_ver='v2', fr_inside_neurons=64,
      fr_stride=1, crop_remove_each=30, value_crop_size=4,
      action_sample_type='sample', action_sample_combine_type='one_or_other',
      sample_gt_prob_type='inverse_sigmoid_decay', dagger_sample_bn_false=True,
      vin_num_iters=36, isd_k=750., use_agent_loc=False, multi_scale=True,
      readout_maps=False, rom_arch=readout_maps_arch_args)

  return arch_args, mapper_arch_args 

# 需要導入模塊: from src import utils [as 別名]
# 或者: from src.utils import Foo [as 別名]
def make_map(padding, resolution, vertex=None, sc=1.):
  """Returns a map structure."""
  min_, max_ = _get_xy_bounding_box(vertex*sc, padding=padding)
  sz = np.ceil((max_ - min_ + 1) / resolution).astype(np.int32)
  max_ = min_ + sz * resolution - 1
  map = utils.Foo(origin=min_, size=sz, max=max_, resolution=resolution,
                  padding=padding)
  return map 

# 需要導入模塊: from src import utils [as 別名]
# 或者: from src.utils import Foo [as 別名]
def reset(self, rngs):
    rng = rngs[0]; rng_perturb = rngs[1];
    nodes = self.task.nodes
    tp = self.task_params

    start_node_ids, goal_node_ids, dists, target_class = \
        _nav_env_reset_helper(tp.type, rng, self.task.nodes, tp.batch_size,
                              self.task.gtG, tp.max_dist, tp.num_steps,
                              tp.num_goals, tp.data_augment,
                              **(self.task.reset_kwargs))

    start_nodes = [tuple(nodes[_,:]) for _ in start_node_ids]
    goal_nodes = [[tuple(nodes[_,:]) for _ in __] for __ in goal_node_ids]
    data_augment = tp.data_augment
    perturbs = _gen_perturbs(rng_perturb, tp.batch_size,
                             (tp.num_steps+1)*tp.num_goals,
                             data_augment.lr_flip, data_augment.delta_angle,
                             data_augment.delta_xy, data_augment.structured)
    perturbs = np.array(perturbs) # batch x steps x 4
    end_perturbs = perturbs[:,-(tp.num_goals):,:]*1 # fixed perturb for the goal.
    perturbs = perturbs[:,:-(tp.num_goals),:]*1

    history = -np.ones((tp.batch_size, tp.num_steps*tp.num_goals), dtype=np.int32)
    self.episode = utils.Foo(
        start_nodes=start_nodes, start_node_ids=start_node_ids,
        goal_nodes=goal_nodes, goal_node_ids=goal_node_ids, dist_to_goal=dists,
        perturbs=perturbs, goal_perturbs=end_perturbs, history=history,
        target_class=target_class, history_frames=[])
    return start_node_ids