def __init__(self,
pretrained_model_name: Optional[str] = None,
cache_dir: Optional[str] = None,
hparams=None):
super().__init__(hparams=hparams)
# Create the underlying encoder
encoder_hparams = dict_fetch(hparams, GPT2Encoder.default_hparams())
self._encoder = GPT2Encoder(pretrained_model_name=pretrained_model_name,
cache_dir=cache_dir,
hparams=encoder_hparams)
# Create a dropout layer
self._dropout_layer = nn.Dropout(self._hparams.dropout)
# Create an additional classification layer if needed
self.num_classes = self._hparams.num_classes
if self.num_classes <= 0:
self._logits_layer = None
else:
logit_kwargs = self._hparams.logit_layer_kwargs
if logit_kwargs is None:
logit_kwargs = {}
elif not isinstance(logit_kwargs, HParams):
raise ValueError("hparams['logit_layer_kwargs'] "
"must be a dict.")
else:
logit_kwargs = logit_kwargs.todict()
if self._hparams.clas_strategy == 'all_time':
self._logits_layer = nn.Linear(
self._encoder.output_size *
self._hparams.max_seq_length,
self.num_classes,
**logit_kwargs)
else:
self._logits_layer = nn.Linear(
self._encoder.output_size, self.num_classes,
**logit_kwargs)
if self._hparams.initializer:
initialize = get_initializer(self._hparams.initializer)
assert initialize is not None
if self._logits_layer is not None:
initialize(self._logits_layer.weight)
if self._logits_layer.bias is not None:
initialize(self._logits_layer.bias)
self.is_binary = (self.num_classes == 1) or \
(self.num_classes <= 0 and
self._hparams.dim == 1)
def default_hparams():
r"""Returns a dictionary of hyperparameters with default values.
.. code-block:: python
{
# (1) Same hyperparameters as in GPT2Encoder
...
# (2) Additional hyperparameters
"num_classes": 2,
"logit_layer_kwargs": None,
"clas_strategy": `cls_time`,
"max_seq_length": None,
"dropout": 0.1,
"name": `gpt2_classifier`
}
Here:
1. Same hyperparameters as in
:class:`~texar.torch.modules.GPT2Encoder`.
See the :meth:`~texar.torch.modules.GPT2Encoder.default_hparams`.
An instance of GPT2Encoder is created for feature extraction.
2. Additional hyperparameters:
`"num_classes"`: int
Number of classes:
- If **> 0**, an additional `Linear`
layer is appended to the encoder to compute the logits over
classes.
- If **<= 0**, no dense layer is appended. The number of
classes is assumed to be the final dense layer size of the
encoder.
`"logit_layer_kwargs"`: dict
Keyword arguments for the logit Dense layer constructor,
except for argument "units" which is set to `num_classes`.
Ignored if no extra logit layer is appended.
`"clas_strategy"`: str
The classification strategy, one of:
- **cls_time**: Sequence-level classification based on the
output of the last time step. Each sequence has a class.
- **all_time**: Sequence-level classification based on
the output of all time steps. Each sequence has a class.
- **time_wise**: Step-wise classification, i.e., make
classification for each time step based on its output.
`"max_seq_length"`: int, optional
Maximum possible length of input sequences. Required if
`clas_strategy` is `all_time`.
`"dropout"`: float
The dropout rate of the GPT2 encoder output.
`"name"`: str
Name of the classifier.
"""
hparams = GPT2Encoder.default_hparams()
hparams.update({
"num_classes": 2,
"logit_layer_kwargs": None,
"clas_strategy": "cls_time",
"max_seq_length": None,
"dropout": 0.1,
"name": "gpt2_classifier"
})
return hparams