Add precomputed_splits toggle (7e936532) · Commits · 钟慕尧 / deepchem

deepchem/molnet/load_function/molnet_loader.py

+13 −1

Original line number	Diff line number	Diff line
		@@ -88,7 +88,8 @@ class _MolnetLoader(object):
		splitter: Union[dc.splits.Splitter, str, None],
		transformer_generators: List[Union[TransformerGenerator, str]],
		tasks: List[str], data_dir: Optional[str],
		save_dir: Optional[str], **kwargs):
		save_dir: Optional[str],
		precomputed_splits: Optional[bool], **kwargs):
		"""Construct an object for loading a dataset.

		Parameters
		@@ -132,6 +133,7 @@ class _MolnetLoader(object):
		self.tasks = list(tasks)
		self.data_dir = data_dir
		self.save_dir = save_dir
		self.precomputed_splits = precomputed_splits
		self.args = kwargs

		def load_dataset(
		@@ -171,6 +173,12 @@ class _MolnetLoader(object):
		if all_dataset is not None:
		return self.tasks, all_dataset, transformers

		# Try to load precomputed splits.

		if self.precomputed_splits is True:
		train, valid, test = self.load_precomputed_splits()
		return (train, valid, test)

		# Create the dataset

		logger.info("About to featurize %s dataset." % name)
		@@ -210,3 +218,7 @@ class _MolnetLoader(object):
		def create_dataset(self) -> Dataset:
		"""Subclasses must implement this to load the dataset."""
		raise NotImplementedError()

		def load_precomputed_splits(self) -> Tuple[Dataset, ...]:
		"""Subclasses must implement this to load precomputed train/test/eval splits."""
		raise NotImplementedError()

+4 −1

Original line number	Diff line number	Diff line
		@@ -33,7 +33,7 @@ class _USPTOLoader(_MolnetLoader):
		self.sep_reagent = sep_reagent
		self.name = 'USPTO_' + subset

		def create_dataset(self) -> Tuple[Dataset, ...]:
		def create_dataset(self) -> Dataset:
		#####INCOMPLETE/INCORRECT: I don't think this is the right way to bypass the splitter!
		if self.subset not in ['MIT', 'STEREO']:
		raise ValueError("Valid Subset names are MIT and STEREO.")
		@@ -94,6 +94,9 @@ class _USPTOLoader(_MolnetLoader):
		#need to figure out how to return the train, test and valid files!
		return (train_file, test_file, valid_file)

		def load_precomputed_splits(self) -> List[Dataset, ...]:
		pass


		def load_uspto(
		featurizer=None,