def normalise_datadict(datadict, cut_to):
"""Take a dictionary of groundtruth and cut all classes to
`cut_to` items. If a class has fewer items, discard it.
Return newdatadict, removed where `removed` is a dictionary
of items in datadict that haven't been added to newdatadict"""
dataset = collections.defaultdict(list)
for r, cls in datadict.items():
dataset[cls].append(r)
newdataset = {}
remaining = {}
for cls, items in dataset.items():
if len(items) > cut_to:
sample = random.sample(items, cut_to)
for i in sample:
newdataset[i] = cls
rest = list(set(items) - set(sample))
for i in rest:
remaining[i] = cls
return newdataset, remaining
def normalise_datadict(datadict, cut_to):
"""Take a dictionary of groundtruth and cut all classes to
`cut_to` items. If a class has fewer items, discard it.
Return newdatadict, removed where `removed` is a dictionary
of items in datadict that haven't been added to newdatadict"""
dataset = collections.defaultdict(list)
for r, cls in datadict.items():
dataset[cls].append(r)
newdataset = {}
remaining = {}
for cls, items in dataset.items():
if len(items) > cut_to:
sample = random.sample(items, cut_to)
for i in sample:
newdataset[i] = cls
rest = list(set(items)-set(sample))
for i in rest:
remaining[i] = cls
return newdataset, remaining