def characterize(self):
bnx_file=BnxFile(self.input_file)
for molecule in bnx_file.parse():
run_id=molecule.run_id
if run_id not in self.dataset_stats:
self.dataset_stats[run_id]={"length": 0.0, "labels": 0.0}
self.dataset_stats[run_id]["length"] += float(molecule.length)
self.dataset_stats[run_id]["labels"] += float(molecule.num_labels)
for run_id in self.dataset_stats:
stats=self.dataset_stats[run_id]
stats["density"] = stats["labels"] / (stats["length"] / (1000 * 100))
def subset(self, criteria, output_file=None): if output_file==None: output_file=self.output_file with open(output_file, "w") as o_file: bnx_file=BnxFile(self.input_file) for header in bnx_file.getHeaders(): o_file.write(header) for molecule in bnx_file.parse(): if criteria(molecule): bnx_file.write(molecule, o_file)
def generateReducedDataset(self, proportion, output_file=None):
if proportion > 1 or proportion < 0:
raise Exception("proportion must be between 1 and 0")
if output_file is None:
output_file=self.input_file + "_" + str(int(proportion*100))
with open(output_file, "w") as o_file:
bnx_file=BnxFile(self.input_file)
headers=bnx_file.getHeaders()
for header in headers:
o_file.write(header)
o_file.write("# Reduced to " + str(proportion*100) + "% of the original\n")
total_length=0.0
molecule_lengths={}
molecule_ids=[]
for molecule in bnx_file.parse():
total_length+=molecule.length
molecule_lengths[molecule.id] = molecule.length
molecule_ids.append(molecule.id)
seed=random.random()
random.seed(seed)
o_file.write("# Random seed for reduction: " + str(seed) + "\n")
target_removed_length=total_length*(1.0-proportion)
removed_length=0
total_molecules=len(molecule_ids)
removed=set()
abridged={}
while removed_length < target_removed_length:
list_index=int(random.random()*total_molecules)
candidate=molecule_ids[list_index]
removed_length+=molecule_lengths[candidate]
del[molecule_ids[list_index]]
total_molecules-=1
if removed_length <= target_removed_length:
removed.add(candidate)
else:
excess_distance=removed_length-target_removed_length
abridged[candidate]=excess_distance
for molecule in bnx_file.parse():
if molecule.id in removed:
continue
if molecule.id in abridged:
molecule.shrink(abridged[molecule.id])
bnx_file.write(molecule, o_file)
def getOutputFileExtension(self): return BnxFile.getExtension()