def make_model(fc_model, transformer, post_model, n_features, dstore):
"""
Helper function for instantiating LTRModel.
:param fc_model: FCModel used as input block
:param transformer: transformer Encoder used as encoder block
:param post_model: parameters dict for OutputModel output block (excluding d_model)
:param n_features: number of input features
:return: LTR model instance
"""
if fc_model:
fc_model = FCModel(**fc_model,
n_features=n_features,
vocab_size=len(dstore.vocab)) # type: ignore
d_model = n_features if not fc_model else fc_model.output_size
if transformer:
transformer = make_transformer(n_features=d_model,
**asdict(transformer,
recurse=False)) # type: ignore
model = LTRModel(fc_model, transformer, OutputLayer(d_model, **post_model))
if dstore.init_from_fasttext:
fc_model.init_from_fasttext(dstore)
# Initialize parameters with Glorot / fan_avg.
for p in model.parameters():
if p.dim() > 1:
nn.init.xavier_uniform_(p)
return model
def make_model(fc_model, transformer, post_model, n_features):
if fc_model:
fc_model = FCModel(**fc_model, n_features=n_features) # type: ignore
d_model = n_features if not fc_model else fc_model.output_size
if transformer:
transformer = make_transformer(n_features=d_model,
**asdict(transformer,
recurse=False)) # type: ignore
model = LTRModel(fc_model, transformer, OutputLayer(d_model, **post_model))
# Initialize parameters with Glorot / fan_avg.
for p in model.parameters():
if p.dim() > 1:
nn.init.xavier_uniform_(p)
return model