("wide_resnet101_2", "WideResNet101_2", "fully_connected"),
("wide_resnet50_2", "WideResNet50_2", "fully_connected"),
]:
cls_config = type(
name + "ModelConfig",
(PyTorchPreTrainedModelConfig, ),
{},
)
cls_context = type(
name + "ModelContext",
(PyTorchPretrainedContext, ),
{},
)
dffml_cls = type(
name + "Model",
(PyTorchPreTrainedModel, ),
{
"CONFIG": cls_config,
"CONTEXT": cls_context,
"PYTORCH_MODEL": model_name,
"LAST_LAYER_TYPE": last_layer_type,
},
)
dffml_cls = entrypoint(model_name)(dffml_cls)
setattr(sys.modules[__name__], cls_config.__qualname__, cls_config)
setattr(sys.modules[__name__], cls_context.__qualname__, cls_context)
setattr(sys.modules[__name__], dffml_cls.__qualname__, dffml_cls)
f"scikit-{entry_point_name}",
),
),
),
"predict": (str, field("Label or the value to be predicted")),
"features": (Features, field("Features to train on")),
},
)
dffml_cls_ctx = type(
name + "ModelContext",
(ScikitContext, ),
{"applicable_features": applicable_features_function},
)
dffml_cls = type(
name + "Model",
(Scikit, ),
{
"CONFIG": dffml_config,
"CONTEXT": dffml_cls_ctx,
"SCIKIT_MODEL": cls,
},
)
# Add the ENTRY_POINT_ORIG_LABEL
dffml_cls = entrypoint(entry_point_name)(dffml_cls)
setattr(sys.modules[__name__], dffml_config.__qualname__, dffml_config)
setattr(sys.modules[__name__], dffml_cls_ctx.__qualname__, dffml_cls_ctx)
setattr(sys.modules[__name__], dffml_cls.__qualname__, dffml_cls)
self.function = self.LOSS(**self.config._asdict())
@classmethod
def load(cls, class_name: str = None):
for name, loss_class in inspect.getmembers(nn, inspect.isclass):
if name.endswith("Loss"):
if name.lower() == class_name:
return getattr(sys.modules[__name__], name + "Function")
for name, loss_class in inspect.getmembers(nn, inspect.isclass):
if name.endswith("Loss"):
cls_config = make_pytorch_config(name + "Config", loss_class)
cls = entrypoint(name.lower())(type(
name + "Function",
(PyTorchLoss, ),
{
"CONFIG": cls_config,
"CONTEXT": {},
"LOSS": loss_class,
},
))
setattr(sys.modules[__name__], cls.__qualname__, cls)
# TODO Currently only the torch.nn module has annotations
# Add PyTorchOptimizer and PyTorchScheduler after the next torch release
# as they are currently adding type annotations for everything