# Copyright 2021 cstsunfu. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import torch
from typing import Dict, List, Set, Callable
from dlk.core.base_module import BaseModule, BaseModuleConfig
from . import decoder_register, decoder_config_register
from dlk.core.modules import module_config_register, module_register
import copy
[docs]@decoder_config_register("linear_crf")
class LinearCRFConfig(BaseModuleConfig):
"""Config for LinearCRF
Config Example:
>>> {
>>> "module@linear": {
>>> "_base": "linear",
>>> },
>>> "module@crf": {
>>> "_base": "crf",
>>> },
>>> "config": {
>>> "input_size": "*@*", // the linear input_size
>>> "output_size": "*@*", // the linear output_size
>>> "reduction": "mean", // crf reduction method
>>> "output_map": {}, //provide_key: output_key
>>> "input_map": {} // required_key: provide_key
>>> },
>>> "_link":{
>>> "config.input_size": ["module@linear.config.input_size"],
>>> "config.output_size": ["module@linear.config.output_size", "module@crf.config.output_size"],
>>> "config.reduction": ["module@crf.config.reduction"],
>>> }
>>> "_name": "linear_crf",
>>> }
"""
def __init__(self, config: Dict):
super(LinearCRFConfig, self).__init__(config)
self.linear_config = config["module@linear"]
self.crf_config = config["module@crf"]
self.post_check(config['config'], used=[
'input_size',
'output_size',
'reduction',
"return_logits",
])
[docs]@decoder_register("linear_crf")
class LinearCRF(BaseModule):
"""use torch.nn.Linear get the emission probability and fit to CRF"""
def __init__(self, config: LinearCRFConfig):
super(LinearCRF, self).__init__(config)
self._provide_keys = {'logits', "predict_seq_label"}
self._required_keys = {'embedding', 'label_ids', 'attention_mask'}
self.config = config
self.linear = module_register.get('linear')(module_config_register.get('linear')(config.linear_config))
self.crf = module_register.get('crf')(module_config_register.get('crf')(config.crf_config))
[docs] def init_weight(self, method: Callable):
"""init the weight of submodules by 'method'
Args:
method: init method
Returns:
None
"""
self.linear.init_weight(method)
self.crf.init_weight(method)
[docs] def forward(self, inputs: Dict[str, torch.Tensor])->Dict[str, torch.Tensor]:
"""do predict, only get the predict labels
Args:
inputs: one mini-batch inputs
Returns:
one mini-batch outputs
"""
return self.predict_step(inputs)
[docs] def predict_step(self, inputs: Dict[str, torch.Tensor])->Dict[str, torch.Tensor]:
"""do predict, only get the predict labels
Args:
inputs: one mini-batch inputs
Returns:
one mini-batch outputs
"""
logits = self.linear(inputs[self.get_input_name('embedding')])
if self._logits_gather.layer_map:
inputs.update(self._logits_gather([logits]))
inputs[self.get_output_name("predict_seq_label")] = self.crf(logits, inputs[self.get_input_name('attention_mask')])
return inputs
[docs] def training_step(self, inputs: Dict[str, torch.Tensor])->Dict[str, torch.Tensor]:
"""do training step, get the crf loss
Args:
inputs: one mini-batch inputs
Returns:
one mini-batch outputs
"""
logits = self.linear(inputs[self.get_input_name('embedding')])
loss = self.crf.training_step(logits, inputs[self.get_input_name('label_ids')], inputs[self.get_input_name('attention_mask')])
if self._logits_gather.layer_map:
inputs.update(self._logits_gather([logits]))
inputs[self.get_output_name('loss')] = loss
return inputs
[docs] def validation_step(self, inputs: Dict[str, torch.Tensor])->Dict[str, torch.Tensor]:
"""do validation step, get the crf loss and the predict labels
Args:
inputs: one mini-batch inputs
Returns:
one mini-batch outputs
"""
logits = self.linear(inputs[self.get_input_name('embedding')])
loss = self.crf.training_step(logits, inputs[self.get_input_name('label_ids')], inputs[self.get_input_name('attention_mask')])
if self._logits_gather.layer_map:
inputs.update(self._logits_gather([logits]))
inputs[self.get_output_name('loss')] = loss
inputs[self.get_output_name("predict_seq_label")] = self.crf(logits, inputs[self.get_input_name('attention_mask')])
return inputs