def testSingleTrainingBatch(self):
batch1, batch2, batch3, meta = (Mock(), Mock(), Mock(), Mock())
self.singleBatchBuilder.build.side_effect = [batch1, batch2, batch3]
self.metaBatchBuilder.build.return_value = meta
train, valid, test = ([Mock()], Mock(), Mock())
dataset = (train, valid, test)
classes = [0]
classNames = ['class']
self.target = BatchBuilder(self.singleBatchBuilder,
self.metaBatchBuilder,
nTrainingBatches=1)
result = self.target.build(dataset, classes, classNames)
self.singleBatchBuilder.build.assert_has_calls(
[call(train, classes),
call(valid, classes),
call(test, classes)])
self.metaBatchBuilder.build.assert_called_with(dataset, classes,
classNames)
self.assertEqual(
{
'data_batch_1': batch1,
'data_batch_2': batch2,
'data_batch_3': batch3,
'batches.meta': meta
}, result)
def testMultipleTrainingBatches(self):
self.singleBatchBuilder.build.side_effect = batches = [
Mock() for i in range(5)
]
self.metaBatchBuilder.build.return_value = meta = Mock()
train = [Mock() for i in range(11)]
valid, test = (Mock(), Mock())
dataset = (train, valid, test)
classes = [0]
classNames = ['class']
self.target = BatchBuilder(self.singleBatchBuilder,
self.metaBatchBuilder,
nTrainingBatches=3)
result = self.target.build(dataset, classes, classNames)
self.singleBatchBuilder.build.assert_has_calls([
call(train[:3], classes),
call(train[3:6], classes),
call(train[6:], classes),
call(valid, classes),
call(test, classes)
])
self.metaBatchBuilder.build.assert_called_with(dataset, classes,
classNames)
self.assertEqual(
{
'data_batch_1': batches[0],
'data_batch_2': batches[1],
'data_batch_3': batches[2],
'data_batch_4': batches[3],
'data_batch_5': batches[4],
'batches.meta': meta
}, result)
def Create(nTrainingBatches = 1):
return ConvnetBatchCreator(BatchBuilder(SingleBatchBuilder(),
MetaBatchBuilder(),
nTrainingBatches),
BatchRepository(FileSystem(),
cPickleSerializer()))
