How to use the batchflow.Config function in batchflow

def _make_layer(self, *args, inputs=None, base_block=BaseConvBlock, **kwargs):
        # each element in `args` is a dict or module: make a sequential out of them
        if args:
            layers = []
            for item in args:
                if isinstance(item, dict):
                    block = item.pop('base_block', None) or item.pop('base', None) or base_block
                    block_args = {'inputs': inputs, **dict(Config(kwargs) + Config(item))}
                    layer = block(**block_args)
                    inputs = layer(inputs)
                    layers.append(layer)
                elif isinstance(item, nn.Module):
                    inputs = item(inputs)
                    layers.append(item)
                else:
                    raise ValueError('Positional arguments of ConvBlock must be either dicts or nn.Modules, \
                                      got instead {}'.format(type(item)))
            return nn.Sequential(*layers)
        # one block only
        return base_block(inputs=inputs, **kwargs)

These defaults can be changed in :meth:`~.TorchModel.build_config` or when calling :meth:`.Pipeline.init_model`.

        Examples
        --------
        .. code-block:: python

            @classmethod
            def default_config(cls):
                config = TorchModel.default_config()
                config['initial_block'] = dict(layout='cnap', filters=16, kernel_size=7, strides=2,
                                               pool_size=3, pool_strides=2)
                config['body/filters'] = 32
                config['head'] = dict(layout='cnadV', dropout_rate=.2)
                return config
        """
        config = Config()
        config['inputs'] = {}
        config['placeholder_batch_size'] = 2

        config['device'] = None
        config['benchmark'] = True
        config['microbatch'] = None
        config['sync_frequency'] = 1

        config['train_steps'] = None
        config['loss'] = None
        config['optimizer'] = 'Adam'
        config['decay'] = None

        config['order'] = ['initial_block', 'body', 'head']
        config['initial_block'] = {}
        config['body'] = {}

def __init__(self, config=None):
        self.full_config = Config(config)
        self.config = Config(config)
        self.train_lock = threading.Lock()

        self.input_names = None
        self.input_shapes = None
        self.target_shape = None
        self.classes = None
        self.model = None
        self.device = None
        self.devices = []
        self.train_steps = None

        self.sync_counter = 0
        self.microbatch = None

        self.iter_info = {}

def __init__(self, config=None):
        self.full_config = Config(config)
        self.config = Config(config)
        self.train_lock = threading.Lock()

        self.input_names = None
        self.input_shapes = None
        self.target_shape = None
        self.classes = None
        self.model = None
        self.device = None
        self.devices = []
        self.train_steps = None

        self.sync_counter = 0
        self.microbatch = None

        self.iter_info = {}

def get_defaults(cls, name, kwargs):
        """ Fill block params from default config and kwargs """
        config = cls.default_config()
        _config = Config(config.get(name))
        _config = _config + (kwargs or {})
        config = {**config['common'], **_config}
        return config

These defaults can be changed in :meth:`~.TorchModel.build_config` or when calling :meth:`.Pipeline.init_model`.

        Examples
        --------
        .. code-block:: python

            @classmethod
            def default_config(cls):
                config = TorchModel.default_config()
                config['initial_block'] = dict(layout='cnap', filters=16, kernel_size=7, strides=2,
                                               pool_size=3, pool_strides=2)
                config['body/filters'] = 32
                config['head'] = dict(layout='cnadV', dropout_rate=.2)
                return config
        """
        config = Config()
        config['inputs'] = {}
        config['placeholder_batch_size'] = 2

        config['device'] = None
        config['benchmark'] = True
        config['microbatch'] = None
        config['sync_frequency'] = 1

        config['train_steps'] = None
        config['loss'] = None
        config['optimizer'] = 'Adam'
        config['decay'] = None

        config['order'] = ['initial_block', 'body', 'head']
        config['initial_block'] = {}
        config['body'] = {}

def config(self):
        """ Returns values. """
        return Config({item[0].value: item[1].value for item in self._config})

These defaults can be changed in :meth:`~.TFModel.build_config` or when calling :meth:`.Pipeline.init_model`.

        Examples
        --------
        .. code-block:: python

            @classmethod
            def default_config(cls):
                config = TFModel.default_config()
                config['initial_block'] = dict(layout='cnap', filters=16, kernel_size=7, strides=2,
                                               pool_size=3, pool_strides=2)
                config['body/filters'] = 32
                config['head'] = dict(layout='cnadV', dropout_rate=.2)
                return config
        """
        config = Config()
        config['inputs'] = {}
        config['initial_block'] = {}
        config['body'] = {}
        config['head'] = {}
        config['predictions'] = None
        config['output'] = None
        config['optimizer'] = ('Adam', dict())
        config['decay'] = (None, dict())
        config['scope'] = ''
        config['common'] = {'batch_norm': {'momentum': .1}}

        return config