def predict(self, x: Tensor, device=None) -> PredictResults:
if self._model is not None:
if hasattr(self._model, "predict_proba"):
proba = torch.as_tensor(
# pyre-fixme[16]: `None` has no attribute `predict_proba`.
self._model.predict_proba(x),
dtype=torch.float,
device=device,
)
score = (proba * torch.arange(proba.shape[1])).sum(dim=1)
return PredictResults(torch.argmax(proba, 1), score, proba)
elif hasattr(self._model, "predict"):
return PredictResults(
None,
torch.as_tensor(
# pyre-fixme[16]: `None` has no attribute `predict`.
self._model.predict(x),
dtype=torch.float,
device=device,
),
None,
)
else:
raise AttributeError(
"model doesn't have predict_proba or predict")
else:
raise Exception("model not trained")
def predict(self, x: Tensor, device=None) -> PredictResults:
if self._model is not None:
self._model.eval()
proba = torch.as_tensor(self._model(x), dtype=torch.float, device=device)
return PredictResults(torch.argmax(proba, 1), proba)
else:
raise Exception("mode not trained")
def predict(self, x: Tensor, device=None) -> PredictResults:
if self._model is not None:
# pyre-fixme[16]: `None` has no attribute `eval`.
self._model.eval()
# pyre-fixme[29]: `None` is not a function.
proba = torch.as_tensor(self._model(x),
dtype=torch.float,
device=device)
return PredictResults(torch.argmax(proba, 1), proba)
else:
raise Exception("mode not trained")