def test_batch_slice_value_cached(self):
'''Test that the values of a batch are correct.
Test the slice funcationality.
'''
array = np.arange(128).reshape((64, 2))
test_array = np.arange(32, 96).reshape((32, 2))
Generator = batch_generator(array, 32, 2, start=16)
equality = np.all(next(Generator) == test_array)
self.assertTrue(equality)
test_array = np.arange(96, 128).reshape((16, 2))
equality = np.all(next(Generator) == test_array)
self.assertTrue(equality)
self.assertRaises(StopIteration, next, Generator)
array = np.arange(64).reshape((32, 2))
test_array = np.arange(32).reshape((16, 2))
Generator = batch_generator(array, 32, 2, end=16)
equality = np.all(next(Generator) == test_array)
self.assertTrue(equality)
self.assertRaises(StopIteration, next, Generator)
array = np.arange(128).reshape((64, 2))
test_array = np.arange(32, 64).reshape((16, 2))
Generator = batch_generator(array, 32, 2, start=16, end=32)
equality = np.all(next(Generator) == test_array)
self.assertTrue(equality)
self.assertRaises(StopIteration, next, Generator)
def test_batch_shape(self):
'''Test that the batch_generator produces predictable batches.'''
array = np.empty((32,1))
self.assertEqual(next(batch_generator(array, 32)).shape, (32,1))
array = np.empty((64,10))
self.assertEqual(next(batch_generator(array, 32)).shape, (32,10))
array = np.empty((16,1))
self.assertEqual(next(batch_generator(array, 32)).shape, (16,1))
array = np.empty((32,1))
self.assertEqual(next(batch_generator(array[:16], 32)).shape, (16,1))
def test_batch_slice_shape_cached(self):
'''Test that the batch_generator with cache produces predictable batches.
Test the slice funcationality
'''
cache_size = 4
array = np.empty((32,1))
self.assertEqual(next(batch_generator(array, 32, cache_size, start=16)).shape, (16,1))
array = np.empty((64,1))
self.assertEqual(next(batch_generator(array, 32, cache_size, start=16, end=32)).shape, (16,1))
def test_batch_shape_cached(self):
'''Test that the batch_generator with cache produces predictable batches.'''
cache_size = 4
array = np.empty((32,1))
self.assertEqual(next(batch_generator(array, 32, cache_size)).shape, (32,1))
array = np.empty((64,1))
self.assertEqual(next(batch_generator(array, 32, cache_size)).shape, (32,1))
array = np.empty((16,1))
self.assertEqual(next(batch_generator(array, 32, cache_size)).shape, (16,1))
array = np.empty((32,1))
self.assertEqual(next(batch_generator(array[:16], 32, cache_size)).shape, (16,1))
def test_batch_value_cached(self):
'''Test that the values of a batch are correct.'''
array = np.arange(128).reshape((64, 2))
test_array = np.arange(64).reshape((32, 2))
Generator = batch_generator(array, 32, 2)
equality = np.all(next(Generator) == test_array)
self.assertTrue(equality)
test_array = np.arange(64, 128).reshape((32, 2))
equality = np.all(next(Generator) == test_array)
self.assertTrue(equality)
self.assertRaises(StopIteration, next, Generator)
array = np.arange(126).reshape((63, 2))
test_array = np.arange(64).reshape((32, 2))
Generator = batch_generator(array, 32, 2)
equality = np.all(next(Generator) == test_array)
self.assertTrue(equality)
test_array = np.arange(64, 126).reshape((31, 2))
equality = np.all(next(Generator) == test_array)
self.assertTrue(equality)
self.assertRaises(StopIteration, next, Generator)
def train_model(model, args, X_train, X_valid, y_train, y_valid):
#Saves the model after every epoch.
checkpoint = ModelCheckpoint('model-{epoch:03d}.h5',
monitor='val_loss',
verbose=0,
save_best_only=args.save_best_only,
mode='auto')
#Use the difference between expected speed and actual speed as a loss function
model.compile(loss='mean_squared_error', optimizer=Adam(lr=args.learning_rate))
#Fits the model on data generated batch-by-batch by a Python generator.
model.fit_generator(batch_generator(args.data_dir, X_train, y_train, args.batch_size, True),
args.samples_per_epoch,
args.nb_epoch,
max_q_size=1,
validation_data=batch_generator(args.data_dir, X_valid, y_valid, args.batch_size, False),
nb_val_samples=len(X_valid),
callbacks=[checkpoint],
verbose=1)