in Model.py [0:0]
def get_conf(layer_id, layer_name, input_layer_ids, all_layer_configs, model_input_ids, use_gpu,
conf_dict=None, shared_conf=None, succeed_embedding_flag=False, output_layer_flag=False,
target_num=None, fixed_lengths=None, target_dict=None):
""" get layer configuration
Args
layer_id: layer identifier
layer_name: name of layer such as BiLSTM
input_layer_ids (list): the inputs of current layer
all_layer_configs (dict): records the conf class of each layer.
model_input_ids (set): the inputs of the model, e.g. ['query', 'passage']
use_gpu:
conf_dict:
shared_conf: if fixed_lengths is not None, the output_dim of shared_conf should be corrected!
flag:
output_layer_flag:
target_num: used for inference the dimension of output space if someone declare a dimension of -1
fixed_lengths
Returns:
configuration class coresponds to the layer
"""
if shared_conf:
conf = copy.deepcopy(shared_conf)
else:
try:
conf_dict['use_gpu'] = use_gpu
# for Embedding layer, add weight_on_gpu parameters
if layer_id == EMBED_LAYER_ID:
conf_dict['weight_on_gpu'] = conf_dict['conf']['weight_on_gpu']
del conf_dict['conf']['weight_on_gpu']
# for classification tasks, we usually add a Linear layer to project the output to dimension of number of classes. If we don't know the #classes, we can use '-1' instead and we would calculate the number of classes from the corpus.
if layer_name == 'Linear':
if isinstance(conf_dict['hidden_dim'], list):
if conf_dict['hidden_dim'][-1] == -1:
assert output_layer_flag is True, "Only in the last layer, hidden_dim == -1 is allowed!"
assert target_num is not None, "Number of targets should be given!"
conf_dict['hidden_dim'][-1] = target_num
elif conf_dict['hidden_dim'][-1] == '#target#':
logging.info('#target# position will be replace by target num: %d' % target_num)
conf_dict['hidden_dim'][-1] = target_num
elif isinstance(conf_dict['hidden_dim'], int) and conf_dict['hidden_dim'] == -1:
assert output_layer_flag is True, "Only in the last layer, hidden_dim == -1 is allowed!"
assert target_num is not None, "Number of targets should be given!"
conf_dict['hidden_dim'] = target_num
elif isinstance(conf_dict['hidden_dim'], str) and conf_dict['hidden_dim'] == '#target#':
logging.info('#target# position will be replace by target num: %d' % target_num)
conf_dict['hidden_dim'] = target_num
# add some necessary attribute for CRF layer
if layer_name == 'CRF':
conf_dict['target_dict'] = target_dict
conf = eval(layer_name + "Conf")(**conf_dict)
except NameError as e:
raise LayerConfigUndefinedError("\"%sConf\" has not been defined" % layer_name)
# verify the rank consistence of joint layers
if layer_name == EMBED_LAYER_NAME:
# the embedding layer
pass
else:
# make sure all the inputs to current layer exist
for input_layer_id in input_layer_ids:
if not (input_layer_id in all_layer_configs or input_layer_id in model_input_ids):
raise ConfigurationError("The input %s of layer %s does not exist. Please define it before "
"defining layer %s!" % (input_layer_id, layer_id, layer_id))
former_output_ranks = [all_layer_configs[input_layer_id].output_rank if input_layer_id in all_layer_configs else all_layer_configs[EMBED_LAYER_ID].output_rank for input_layer_id in input_layer_ids]
# inference input_dim
conf.input_dims = [all_layer_configs[input_layer_id].output_dim if input_layer_id in all_layer_configs else all_layer_configs[EMBED_LAYER_ID].output_dim for input_layer_id in input_layer_ids]
# If the inputs come from embedding layer and fixed_lengths exist, set the length to input_dims
if len(input_layer_ids) == 1 and input_layer_ids[0] in model_input_ids and fixed_lengths:
conf.input_dims[0][1] = fixed_lengths[input_layer_ids[0]]
# check and verify input ranks
if conf.num_of_inputs > 0:
if conf.num_of_inputs != len(input_layer_ids):
raise ConfigurationError("%s only accept %d inputs but you feed %d inputs to it!" % \
(layer_name, conf.num_of_inputs, len(input_layer_ids)))
elif conf.num_of_inputs == -1:
conf.num_of_inputs = len(input_layer_ids)
if isinstance(conf.input_ranks, list):
conf.input_ranks = conf.input_ranks * conf.num_of_inputs
else:
logging.warning("[For developer of %s] The input_ranks attribute should be a list!" % (layer_name))
[conf.input_ranks] * conf.num_of_inputs
for input_rank, former_output_rank in zip(conf.input_ranks, former_output_ranks):
if input_rank != -1 and input_rank != former_output_rank:
raise ConfigurationError("Input ranks of %s are inconsistent with former layers" % layer_id)
conf.input_ranks = copy.deepcopy(former_output_ranks)
# inference and varification inside the layer
conf.inference() # update some attributes which relies on input dimension or something else
conf.verify() # verify if the configuration is legal
former_conf = None if len(all_layer_configs) == 0 else list(all_layer_configs.values())[-1]
conf.verify_former_block(former_conf) # check if has special attribute rely on former layer
logging.debug('Layer id: %s; name: %s; input_dims: %s; input_ranks: %s; output_dim: %s; output_rank: %s' % (layer_id, layer_name, conf.input_dims if layer_id != 'embedding' else 'None', conf.input_ranks, conf.output_dim, conf.output_rank))
return conf